The Influence of CEO Narcissism and Audit Committee Status on Audit Risk
MSc Accountancy & Controlling
Combined MSc Thesis Accountancy (EBM869B20) & Controlling (EBM870B20) University of Groningen, Faculty of Economics and Business
WILLEM SCHOLTEN S2339676 Van Leijenberghlaan 153 1082 GE Amsterdam +31 6 10 40 69 46 w.h.scholten@student.rug.nl
Supervisors: Prof. Dr. J.A. Emanuels Assessor: Dr. Y. Karaibrahimoglu Co-assessor: G.C. Helminck RA MSc EMA
2
TABLE OF CONTENTS
Abstract ... 3
1. INTRODUCTION ... 4
1.1 Contribution ... 6
2. THEORETICAL BACKGROUND AND HYPOTHESIS DEVELOPMENT ... 8
2.1 CEO Narcissism and Audit Risk ... 8
2.2 CEO narcissism and Audit Committee Status ... 9
2.3 Audit Committee Status and Audit Risk ... 10
2.3.1 Agency and Resource Dependence Theory ... 11
2.3.1 Status Characteristics Theory ... 12
3. RESEARCH METHODOLOGY ... 15
3.1 Sample Selection ... 15
3.2 Regression Model and Variables... 15
3.2.1 Dependent variable – Audit Risk ... 17
3.2.2 Independent variable – CEO Narcissism ... 18
3.2.3 Mediating variable – Audit Committee Status ... 18
3.2.4 Moderating Variable – Audit Committee Size ... 19
3.2.5 Control variables... 20 4. RESULTS ... 22 4.1 Descriptive Statistics ... 22 4.2 Correlation Matrix ... 24 4.3 Main analysis... 25 4.4 Additional analysis ... 27
4.4.1 Structural equation model ... 27
4.4.2 Linear regression analysis ... 30
5. CONCLUSION AND DISCUSSION ... 33
5.1 Limitations and future research possibilities ... 35
3 Abstract
In this research, theoretical insights from the Upper Echelons Theory were used to investigate the relationship between CEO narcissism and audit risk. Furthermore, the mediating effect of audit committee status and the moderating role of audit committee size on the mediating relationship was investigated. The initial sample of this research consisted of 537 listed firms from the UK and the Netherlands over the period 2010-2016. In 2013, the long-form audit report became mandatory, which led to an increase in audit risk-related data. Therefore, audit risk data was collected over the period 2013-2016. This research uses hand-collected data regarding the CEO narcissism and audit risk variables. After omitting data due to CEO changes or data unavailability, the remaining sample comprised 849 firm-year observations. The results point out that CEO narcissism is significantly related to audit risk. However, no statistical evidence was found that audit committee status mediates this relationship. Furthermore, there was no moderated mediation relationship, with audit committee size as moderator. The size of an audit committee did, however, strengthen the relationship between audit committee status and audit risk.
4
1. INTRODUCTION
In the recent past, the Big 4 accountancy firms have received a lot of media attention due to accounting scandals. For example, in August 2017, KPMG was fined by the SEC, because they gave an unqualified opinion for a subsidiary of an oil and gas company, Miller Energy Resources, which overvalued certain assets by more than 100 times (Financial Times, 2017). In their official statement, the SEC stated that KPMG failed their audit because they lacked to make a proper risk assessment for taking Miller Energy Resources as a client and did not have the appropriate staff available for the audit (SEC, 2017). In such a risk assessment, the auditor takes into account several client-specific risks, such as business risks and fraud risks, which can impact the overall audit risk assessment (Johnstone, 2000). Furthermore, they stated that the accountancy firm should have established policies and procedures for accepting or continuing the engagement with the client, and that KPMG failed to do so. In the end, not only KPMG was fined, but the SEC has also settled charges for accounting fraud with both the CEO and the CFO of the subsidiary (Financial Times, 2017).
These types of scandals have occurred in all Big 4 accountancy firms, resulting in questions about the value of the auditor and putting more emphasis on the importance of a proper risk-based approach in the audit process. Therefore, in this research, we will focus on the audit risk, perceived by the auditor. We will conduct this research from the upper echelons point of view, since we want to investigate the effect of personal characteristics of top managers in this relationship, as they are held responsible for organizational outcomes like accounting fraud, as mentioned in the above paragraph.
Upper echelons theory, developed by Hambrick and Mason (1984), views organizations as a reflection of their top managers. More precisely, organizations are influenced by the cognitive base, personal values and choices of powerful actors in the organization. There has been extensive research into several personality traits of top management team (TMT) members and their influence on organizational level outcomes, for example, the relationship between charismatic leadership and firm performance (Waldman et al., 2004); the relationship between the business-related background of higher level managers and more sophisticated management accounting and control systems (Hiebl, 2014); and the relationship between TMT tenure and strategic persistence (Finkelstein and Hambrick, 1990). Here, we will focus on how narcissism of the CEO as a personality trait can influence organizational level outcomes. Prior literature showed that CEO
5
narcissism is, for example, related to financial misreporting (Hales et al., 2011). This could have been the case in the Miller Energy Resources scandal, mentioned above.
The concept of narcissism has been extensively researched since the early twentieth century, starting with work from, for instance, Freud (1914). Nowadays, there are several definitions of narcissism. Zhu and Chen (2015, p. 32) summarize the definitions of several authors as follows: “the degree to which an individual has an inflated self-view and craves affirmation of that self-view”. They also recognize that narcissism is a fundamental personality dimension for CEOs that can impact organizational outcomes, which is in line with the upper echelons theory. Existing literature shows that CEO narcissism is related to, for example, audit fees, excessive risk taking, the auditor’s overall fraud risk assessment (Judd et al., 2017; Buyl et al., 2017; Johnson et al., 2012) and several other negative organizational outcomes like earnings management and financial misreporting (Ham et al., 2017; Hales et al., 2011). Houston et al. (1999) argue that the auditor also respond to these detrimental organizational outcomes in assessing the overall risk for the engagement with the client. Thus, we will investigate whether and how CEO narcissism influences the audit risk of a firm, perceived by the auditor.
Furthermore, narcissism is associated with self-focus and admiration and dominance (Zhu & Chen, 2015; Chatterjee & Pollock, 2017). However, according to Tang et al. (2011), the effects of dominance of the narcissistic CEO are weakened by a powerful board. Chatterjee and Pollock (2017) furthermore propose a relationship between CEO narcissism and the status of board members. Here, we specifically focus on the audit committee, as a subgroup of the board. Thus, we will also look at how the audit committee exerts influence on the CEO. This could be explained from two theoretical perspectives. First, a combination of the agency and resource dependence theory, in which the principal and agent have different goals or risk attitudes and act in self-interest and, therefore, the agent should be monitored to ensure that the agent behaves in the principal’s interests (Eisenhardt, 1989). In this case, the CEO is the agent and the audit committee monitors the CEO on behalf of the shareholders (i.e., the principal). Furthermore, the audit committee provides resources to the firm (Hillman & Dalziel, 2003). High status audit committee members are likely to perform their tasks better than others, because of their expertise and experience and because they are less deferential to the CEO than lower status audit committees (Badolato et al., 2014). According to the SEC (2003b), this results in more active monitoring of the CEO and providing more resources to the firm. Second, the status characteristic theory explains that
6
individuals’ behavior differs among groups, based on the status they have in that certain group (Anderson et al., 2006). This could indicate that when the audit committee members have a higher status than the CEO, the CEO might conform to the audit committee.
Since there are some inconsistencies in terminology regarding the board, especially in the Dutch sample where executives and non-executives are separated, we think this should be clarified in order to fully understand this research. On the one hand, we use the term ‘board’ for non-executives. Therefore, the audit committee is a subgroup of the board. On the other hand, if necessary, we use the term ‘executives’ or ‘Top Management Team’ (TMT) for executives.
In this research, we aim to provide empirical evidence on whether the personality trait CEO narcissism influences the degree of audit risk, in a sample of listed firms in the Netherlands and UK. Furthermore, we will investigate whether the status of the audit committee and the number of audit committee members interfere in this relationship. Thus, the research question is as follows:
Does audit committee status mediate the relationship between CEO narcissism and is this relationship moderated by the size of the audit committee?
1.1 Contribution
In this research, we make several contributions to literature and practice. First, in existing literature, there has been done little research to the relationship between CEO narcissism and audit risk (e.g., Judd et al., 2017). Therefore, we aim to find additional empirical evidence regarding this relationship, with a sample of listed firms in the Netherlands and UK in the fiscal years 2013 - 2016. We contribute to upper echelons theory, by providing evidence regarding the relationship between a personality trait of a powerful actor, the degree of narcissism of the CEO, and a firm-level outcome, audit risk, perceived by the auditor. Also, in providing empirical evidence for this relationship, we used hand-collected data for the CEO narcissism variable. This increases the validity of the research and the availability of data for future research regarding CEO narcissism.
Second, we contribute to literature by extending prior findings by the incorporation of audit committee status in the predicted model. This has not yet been investigated. In this model, we aim to investigate whether audit committee status is a mediator in the relationship between CEO narcissism and audit risk. Here, we provide evidence regarding the proposed relationship between CEO narcissism and audit committee status, as developed, but not tested by Chatterjee and Pollock (2017); Specifically, we investigate whether audit committee status mediates the relationship between CEO narcissism and audit risk.
7
Third, in existing literature, there are mixed results about the relationship between audit committee characteristics and audit risk (Goodwin-Steward & Kent, 2006; Abbott et al., 2003; Abbott et al., 2004)). Therefore, we aim to provide empirical evidence regarding the effect of audit committee status and the number of audit committee members on audit risk.
Finally, in prior literature, the most commonly used proxy for audit risk was audit fees (Brockman et al., 2013; Hogan & Wilkins, 2008). In this research, we use a direct measure for audit risk, with hand-collected data from annual reports and audit reports. This will be directly measured with the areas of focus, or key audit matters, that could lead to material misstatements in the financial statement, assessed by the auditor. This data has increased in availability since the mandatory implementation of the long-form audit report in 2013, in which these risks are disclosed.
The results are also relevant for practice. In particular, the external auditor and the organization itself can learn from the provided statistical evidence. For example, when the auditor indicates that their client has a narcissistic CEO and a high status audit committee, this may lead to a either a higher or lower audit risk. In that case, the auditor has to perform more or less substantive procedures, respectively (Hogan & Wilkins, 2008). Also, according to Johnstone (2000), the auditor uses the degree of audit risk in the negotiations with the client regarding the audit fees. This implicates higher audit fees when the audit risk increases. Furthermore, the results could be relevant for the organization itself. Johnstone (2000) argues that auditors use their audit risk assessment in their client acceptance process. When we apply this to the relationship between CEO narcissism and audit risk, with audit committee status as a mediator, this means that when a narcissistic CEO results in a either a higher or lower audit risk, the chance that the auditor accepts the client will decrease or increase, respectively. Furthermore, when audit committee size moderates this relationship, the effects can be strengthened when there is a large audit committee.
In the following section we will provide the theoretical backgrounds, which lead to the hypotheses. Afterwards, the research methodology will be discussed, followed by the results, discussion of the findings and conclusion.
8
2. THEORETICAL BACKGROUND AND HYPOTHESIS DEVELOPMENT
In this section the concepts of CEO narcissism, audit risk and audit committee status and members will be discussed. Furthermore, this sections contains the hypotheses that will be tested.
2.1 CEO Narcissism and Audit Risk
Upper echelons theory explains why organizations do what they do, as organizations are reflections of powerful actors in the organization (Hambrick & Mason, 1984). This means that personal characteristics or values of top management team (TMT) members can affect organizational outcomes or attributes, like firm performance and, for example, tone at the top (Judd et al., 2017). The authors here stress that the tone at the top, as a reflection of the CEO’s personal characteristics, can influence the inherent and control risk, assessed by the auditor. The focus of this research is on the effect of narcissism, as a characteristic of the CEO, on an organizational level outcome: audit risk, perceived by the auditor.
According to Zhu and Chen (2015), a narcissist has an inflated self-view and searches for affirmation of the inflated self-view by others. Campbell et al. (2000) state that narcissism leads to positivity, egocentrism and a sense of uniqueness. Findings of Williams (2014) indicate that the relationship between power and the pursue of interested behavior is strengthened by self-focused goals, which can arise from personality traits like narcissism. These findings are confirmed in existing literature, in which is elicited that CEO narcissism relates to detrimental organizational outcomes as fraud and financial misreporting (Hales et al., 2011), earnings management (Ham et al., 2017), and internal control weaknesses (Judd et al., 2017). This could indicate an increase in audit risk for a firm with a narcissistic CEO.
Audit risk encompasses the risk that an auditor issues an unqualified opinion on financial statements that contains a material misstatement (Houston et al., 1999) and it can be calculated using the Audit Risk Model (ARM), developed by the American Institute of Certified Public Accountants (AICPA) in Statement on Auditing Standards (SAS) No. 47 (1983). According to the Audit Risk Model, the audit risk is calculated as follows:
9
Thus, the audit risk consists of three elements: inherent risk (IR), control risk (CR) and detection risk (DR). According to Hogan and Wilkins (2008), the auditor assesses the inherent risk and control risk based on a client assessment. When the auditor discovers a high inherent risk and control risk, the auditor must decrease the detection risk by increasing the substantive testing. All three types of risk that assess the audit risk are related to the probability that there is a material misstatement in the financial statements, they can be defined as follows: IR is the risk that exist before considering the effectiveness of internal controls, CR is the probability that a material misstatement is not resolved by the internal controls, and DR is the tolerable level of risk that auditing procedures will not detect material misstatements (Houston et al., 1999). Thus, DR can be changed by increasing or decreasing substantive testing, to keep the audit risk within the acceptable level, set by the auditor. Therefore, to keep the audit risk within that level, the amount of substantive testing is dependent on IR and CR, as those risks are assessed by the auditor and cannot be increased or decreased by the amount of substantive procedures.
As indicated in prior research, CEO narcissism can impose several business risks, like earnings management (Ham et al., 2017) and financial reporting fraud (Hales et al., 2011). This means that there is an opportunity to commit fraud or engage in earnings management and, following the ARM, that the internal controls in place are not completely effective. Furthermore, Judd et al. (2017) state that narcissistic CEOs are likely to pose greater inherent risk and control risk, since they feel the need to provide a more positive view of themselves and their firms. Also, the authors add that narcissists feel that they are above the law and, therefore, do not always feel the need to comply with rules and regulations.
2.2 CEO narcissism and Audit Committee Status
Chatterjee and Pollock (2017) argue that existing literature only considers the direct relationship between CEO narcissism and organizational-level outcomes. In other words, the authors state that there is a lack of theorizing about mediators or other mechanisms that explain those relationships. Thus, the authors use the upper echelons theory to explain organizational outcomes, but with an emphasis on the mediating structures that explain why CEO narcissism is related to these outcomes. Chatterjee and Pollock (2017) look at the way how CEOs structure their organizations or external stakeholders. This resulted in a framework they have elaborated, but not empirically tested. In this framework, Chatterjee and Pollock (2017) explain how narcissistic
10
CEOs cope with two conflicting needs: (1) the need for acclaim and (2) the need to dominate others.
The need to be acclaimed by others can be fulfilled by being associated with other people with a high status, since the authors argue that interaction with high status individuals can enhance your own status. Status is “the ability to influence outcomes based on perceived skills, qualities and personal attributes” (Badolato et al., 2014, p. 208). Besides that narcissists want to enhance their own status, Chatterjee and Pollock (2017) also argue that narcissistic CEOs want to dominate others in decision-making, and dominated individuals are not likely to acclaim the dominator. Therefore, Chatterjee and Pollock (2017) argue that narcissistic CEOs structure their TMTs and boards in a way that they can both dominate others and fulfill their need to be acclaimed by others. Thus, they try to dominate the daily decision making by appointing lower status TMT members, and fulfill their need for acclaim by appointing high status board members (Chatterjee and Pollock, 2017). This line of reasoning is supported by the article of Campbell et al. (2004), in which the authors state that narcissistic people, on the one hand, gain esteem by associating with high-status others. On the other hand, they score high on agentic traits like dominance. To conclude, we expect that narcissistic CEOs are likely to appoint higher status board members and, therefore, higher status audit committee members, in order to gain esteem and fulfill their need to be acclaimed by others. How this affects the audit risk will be discussed in the next section.
2.3 Audit Committee Status and Audit Risk
Existing literature gives us several insights on how audit committee characteristics can influence the audit risk. However, these insights have been found to be contradictory. The articles of Abbott et al. (2003) and Abbott et al. (2004), for example, present findings that can be interpreted in different ways with respect to audit risk. Where Abbott et al. (2003) indicate that audit committee independence is related to higher audit fees, Abbott et al. (2004) state that audit committee independence is related to less restatements. When using the Audit Risk Model in interpreting these results, the following can be deduced from these articles. First, higher audit fees indicate a higher audit risk because, as mentioned above, audit fees increase when the auditor has to perform more substantive procedures, which are costly (Hogan & Wilkins, 2008). Second, restatements can imply a weakness in internal controls of a firm (Doyle et al., 2007), therefore, lower restatements can be related to a lower control risk. A lower control risk, in turn, relates to a
11
lower audit risk, ceteris paribus. In the following section, we will provide different insights on how audit committee status can influence audit risk, to develop our hypothesis.
Here, we discuss the role of status of the audit committee status from two theoretical perspectives: (1) a combination of the agency theory and resource dependence theory, used by Hillman and Dalziel (2003); and (2) the status characteristics theory, as used by Veltrop et al. (2017). These perspectives give us several insights on why – and the direction in which – audit risk is affected by audit committee status.
2.3.1 Agency and Resource Dependence Theory
The function of the audit committee in this relationship can be derived from the definition of Audit Committee Effectiveness (ACE), provided by DeZoort et al. (2002, p. 41): “An effective
audit committee has qualified members with the authority and resources to protect stakeholder interests by ensuring reliable financial reporting, internal controls, and risk management through its diligent oversight efforts.” This definition shows both functions of an audit committee,
according to Hillman and Dalziel (2003): (1) the monitoring function to protect shareholders interests, and (2) the provision of resources to the board.
According to Hillman and Dalziel (2003), the monitoring function stems from the agency theory, where the agent should be monitored on behalf of the principal, in a situation of separated ownership, which leads to information asymmetry. In this research, the CEO is the agent and the shareholders are the principals. Providing resources is the second function, and stems from the resource dependence theory. This refers to the ability to provide resources to the firm, for example, advice and counsel or communication channels with external organizations (Hillman & Dalziel, 2003).
Badolato et al. (2014) discuss how status can influence the execution of these functions in two ways. First, members with higher relative status will be seen as individuals with higher authority and more competent, therefore, they are likely to be able to provide better resources. Second, higher status audit committee members are perceived to be active monitors. Since they want to keep their high status, they perform their function well, so that they minimize the threat that they will be associated with failures, which can diminish their status.
However, these two functions have different implications with respect to the audit risk. The monitoring function, on the one hand, can lead to the identification of new risks associated with
12
the audit. When the audit committee has identified these risks, the findings should be reviewed with the external auditor (Smith Report, 2003). In turn, the auditor can use these risks in their own work, and record them in the independent auditor’s report. We believe that high status audit committees are likely to be more diligent. When an audit committee is diligent in the execution of their monitoring, they are assumed to identify more audit risks (Goodwin-Stewart & Kent, 2006).
On the other hand, we believe that the function of providing resources can lead to a decrease in the number of audit risks. One type of resource that audit committees can provide is advice and counsel (Hillman & Dalziel, 2003). Thus, the audit committee advices the TMT of a firm. When the audit committee gives advice regarding the strengthening of internal controls within the firm, and the TMT follows this advice, the control risk decreases, resulting in a lower audit risk. In this case also applies that high status audit committee members are more likely to be active in the execution of their function. Therefore, we believe that high status audit committee members can also lead to a decrease in audit risk.
Abbott et al. (2003) support this point view, by arguing that there are two ways to look at the relationship between audit committee characteristics and audit fees. They argue that stronger audit committees can be both associated with higher and lower audit fees. As explained, on the one hand, high status audit committees can be more diligent and identify more audit related risks. On the other hand, an effective audit committee can also reduce the number of internal control weaknesses, and therefore, reduce the perceived audit risk (Abbott et al., 2013). Thus, the effects of audit committee characteristics on audit risk can be two sided.
2.3.1 Status Characteristics Theory
According to Anderson et al. (2006), the status characteristics theory explains that interactions and influence of individuals in groups differ because of an individual’s ascribed status. Following this theory, Veltrop et al. (2017) discuss that financial expertise is an important indicator of an individual’s status in a board. Since audit committee members are expected to have a high degree of financial expertise, they are likely to have a higher status within a board. Furthermore, Veltrop et al. (2017) point out that when individuals have a higher status, others are more likely to conform to that individual. This could indicate that the CEO is more likely to be constricted by high status audit committee members. This is in a way supported by the research of Badolato et al. (2014), in which is stated that audit committee status is negatively related to earnings management.
13
In other words, the CEO is constricted by a high status audit committee. Furthermore, Abbott et al. (2004) discuss that at least one audit committee member with high financial expertise is negatively related to audit fees. Following the Audit Risk Model, this means that the audit risk and, therefore, the audit fee is lower for these firms, as the auditor had to perform less substantive procedures. According to Krishnan and Visvanathan (2009), auditors also value financial expertise in the audit committee in their overall audit risk assessment. The authors argue that when the audit committee members have financial expertise, the control risk, assessed by the auditor, decreases. Therefore, following the Audit Risk Model, the overall audit risk decreases. Since financial expertise is an important indicator of status (Veltrop et al., 2017), we investigate whether the audit risks also decreases when audit committees have a high status.
Taking the above into account, from the point of view of the status characteristics theory, we can gained insights on how audit committee status can decrease the audit risk. However, a narcissistic CEO is not likely to be deferential to others with high status, since the CEO believes that his or her status has enhanced, by affiliating with other high status individuals (Chatterjee & Pollock, 2017). Therefore, following the research of Veltrop et al. (2017), a narcissistic CEO, who has an inflated self-view, is less likely to conform or listen to others. This could indicate that the CEO is less likely to follow advice from the audit committee, which the audit committee provides, as seen in the section above.
Although we believe that the status characteristics theory perspective is less likely to hold in situations with a narcissistic CEO, we cannot be conclusive in our hypothesis and rely on solely on the agency and resource dependence theory perspective. Thus, the effect of audit committee status is two sided, which will be reflected in the hypothesis below.
Now, we have discussed why and the directions in which audit committee status can influence the audit risk. Here, we will discuss an additional condition when this will apply. Abbott et al. (2004) argue that audit committees should have sufficient resources to cope with the complexity that comes with the tasks of the audit committee. Thus, the size of an audit committee increases the effectiveness and diligence. This is supported by Kalbers and Fogarty (1993), who state that larger audit committees are more likely to be seen as an authoritative body by external auditors. Therefore, we believe that the size of an audit committee moderates the degree to which audit committee status influences the number of audit risks. As we have discussed above,
14
effectiveness and diligence of an audit committee can either have positive or negative effects on the audit risk. Therefore, we expect a strengthening moderating effect. This means, on the one hand, that when audit committee status results in a low number of audit risks, this number will be even lower when the audit committee is larger. On the other hand, when the audit committee results in a higher number of audit risks, this number will even further increase when the size of the audit committee increases.
Thus, combining the above, we posit that status of audit committee members mediates the relationship between CEO narcissism and audit risk, in a way that it can either increase or decrease the audit risk. Furthermore, we believe that the size of audit committees moderates this mediating relationship. These predictions are summarized in the following hypotheses and in the conceptual model in figure 1.
Hypothesis 1: Audit committee status mediates the relationship between CEO narcissism
and audit risk.
Hypothesis 2: The mediating role of audit committee status in the relationship between
CEO narcissism and audit risk is moderated by audit committee size.
15
3. RESEARCH METHODOLOGY
This section describes the design of the research. Here, we will discuss the sample, the regression models and all variables, including control variables.
3.1 Sample Selection
To conduct this research, we will use a sample of non-financial Dutch and UK-listed firms between the fiscal years 2013 – 2016. This sample and period is particularly interesting to investigate, since the extended audit report became mandatory in 2013 for listed firms (Reid et al., 2016). Therefore, there is more information disclosed in the independent auditor’s report, for example, the materiality and key audit matters. From this sample, both archival and hand-collected data will be used to measure the necessary variables and to provide statistical evidence for the proposed relationships. Archival data will be retrieved from several databases, whereas the hand-collected data will be found in the annual and audit reports of the firms in the sample.
The initial sample consists of 537 companies. To compute the CEO narcissism variable, data was collected from 2010 for UK companies and 2011 for Dutch companies, respectively, until the mandatory implementation of the long-form audit report. Furthermore, audit risk data was collected from the fiscal year ending in 2013 until 2016. This resulted in a total of 3628 observations. Due to CEO changes during the period, unavailability of data and omitting the years 2010 - 2012, the sample has been reduced to 849 firm-year observations.
3.2 Regression Model and Variables
In order to test the hypotheses stated above and answer the research question, we conducted a structural equation model analysis. Therefore, we have developed the following regression models:
Model 1: 𝑅𝑅𝐼𝐼𝑅𝑅𝑅𝑅 = 𝛼𝛼 + 𝛽𝛽1 ∗ 𝑁𝑁𝐴𝐴𝑅𝑅𝐶𝐶𝐼𝐼𝑅𝑅𝑅𝑅𝐼𝐼𝑅𝑅𝑁𝑁 + 𝛽𝛽2 ∗ 𝑅𝑅𝑆𝑆𝐴𝐴𝑆𝑆𝑆𝑆𝑅𝑅 + 𝛽𝛽3 ∗ 𝐶𝐶𝐶𝐶𝑁𝑁𝑆𝑆𝑅𝑅𝐶𝐶𝐶𝐶𝑅𝑅 + 𝜀𝜀 Model 2: 𝑅𝑅𝐼𝐼𝑅𝑅𝑅𝑅 = 𝛼𝛼 + 𝛽𝛽1 ∗ 𝑁𝑁𝐴𝐴𝑅𝑅𝐶𝐶𝐼𝐼𝑅𝑅𝑅𝑅𝐼𝐼𝑅𝑅𝑁𝑁 + 𝛽𝛽2 ∗ 𝑅𝑅𝑆𝑆𝐴𝐴𝑆𝑆𝑆𝑆𝑅𝑅 + 𝛽𝛽3 ∗ (𝐴𝐴𝐶𝐶𝑅𝑅𝐼𝐼𝐴𝐴𝐴𝐴 ∗ 𝑅𝑅𝑆𝑆𝐴𝐴𝑆𝑆𝑆𝑆𝑅𝑅)
+ 𝛽𝛽4 ∗ 𝐶𝐶𝐶𝐶𝑁𝑁𝑆𝑆𝑅𝑅𝐶𝐶𝐶𝐶𝑅𝑅 + 𝜀𝜀
In the following paragraphs, the variables - including control variables - summarized in Table 1 are explained in further detail.
16
Variable Description
RISK Audit risk, measured as the number of risks disclosed by the auditor in the independent auditor’s report.
NARCISSISM 1
The degree of narcissism of the CEO in a firm, measured by a construct consisting of (1) the prominence of the CEO’s photograph in the company’s annual report; (2) the CEO’s use of first-person singular pronouns in interviews; (3) the CEO’s relative cash compensation; and (4) the CEO’s relative non-cash compensation.
NARCISSISM 2
The degree of narcissism of the CEO in a firm, measured by a construct consisting of (1) the prominence of the CEO’s photograph in the company’s annual report; (2) the CEO’s use of first-person singular pronouns in interviews.
STATUS
Status of the audit committee, measured with the number of (1) contemporaneous public board directorships; (2) contemporaneous private board directorships; and (3) degrees and qualifications.
ACSIZE Audit committee size, measured as total number of audit committee members. Control Variables
Year-fixed effects Year-fixed effects, based on financial reporting year. Industry-fixed effects Industry-fixed effects, based on SIC codes.
Country fixed effects Country-fixed effects, based on ISO codes.
Audit company-fixed effects Audit company-fixed effects, divided into Deloitte, EY, KPMG, PwC and Other. LnSIZE Firm size, measured as the log of total assets in euros.
LMV Log of the market value of equity.
ROA Return on assets before extraordinary items.
LEV Leverage, measured as total liabilities over total assets.
NROA Negative return on assets, measured as 1 if the firm has a negative ROA, 0 otherwise. IHRISK Inherent risk, measured as total inventories and receivables over total assets.
AUDFEES% Audit fees, measured as total audit fees over total assets. CEOAGE Age of the CEO, measured in years.
CEOTEN CEO tenure, measured in years.
CEOGEN Gender of the CEO: 1 if the CEO is a male, 0 if the CEO is a female. BSIZE Board size, measured as total number of board members.
17
3.2.1 Dependent variable – Audit Risk
The dependent variable in this research is the audit risk (RISK) of a firm. This will be directly measured as the number of audit risks, disclosed by the auditor in the independent auditor’s report, also known as the areas of focus of key audit matters. This data became available since the implementation of the long-form audit report, which became mandatory since 2013. Since this information has only recently become available, there has been done little research in which this direct measure is used.
In prior research, audit fees were used as a proxy for audit risk. The logic behind this can be derived from the Audit Risk Model, on which we have elaborated in the theory section. As mentioned before, the detection risk is dependent on the inherent and control risk. When the inherent and control risk are high, detection risk should be decreased by increasing substantive procedures, to keep the audit risk within the acceptable level (Hogan & Wilkins, 2008). As the level of substantive procedures increases, the audit fees will also increase, otherwise, it will be too costly for the auditor. However, this might not always be the case. For instance, when fixed fee contracts are used or, as Munsif et al. (2011, p. 104) state: when audit fees are “sticky”. This means that auditors charge a risk premium for firms that have had material weaknesses in their controls in the past, even though they have remediated these weaknesses. In this case, the audit fees do not reflect the amount of substantive procedures that the auditor has performed. Therefore, we believe that audit fees is an imperfect measure for audit risk. Our direct measure will be obtained from the independent auditor’s report, disclosed in the annual reports of our sample firms.
BIND Board independence, measured as number of independent board members over total board members.
BMEET Board meetings, measured as total number of annual board meetings.
ACIND Audit committee independence, measured as number of independent audit committee members over total audit committee members.
LnMAT Materiality, measured as the log of the materiality assessed by the auditor.
INOWN Inside ownership: measured as the cumulative percentage of voting control held by managers and directors.
BLOCK Block holder ownership: measured as the cumulative percentage of voting control held by block holders.
18
3.2.2 Independent variable – CEO Narcissism
The independent variable is CEO narcissism (NARCISSISM), which will be measured by using a construct from Chatterjee and Hambrick (2007). They used this construct, because the prior measure, a survey regarding the Narcissistic Personality Inventory (NPI), was not suitable in this research area. Chatterjee and Hambrick (2007) argued that narcissists are likely to be reluctant to participate in such survey research and that their response will be biased due to the social desirability of their answers. Therefore, the authors developed an unobtrusive measure, which consists of the following indicators for CEO narcissism: (1) the prominence of the CEO’s photograph in the company’s annual report; (2) the CEO’s use of first-person singular pronouns in interviews; (3) the CEO’s relative cash compensation; (4) the CEO’s relative non-cash compensation, comparable to the measure from Chatterjee and Hambrick (2007). A narcissism score will be computed from these four indicator variables. This data will be collected from archival data and annual reports. Using at least one hand-collected variable increases the validity of the research.
However, due to the unavailability of data regarding the CEO’s compensation, we include a second measure, one that excludes the compensation data. Therefore, our first measure (NARCISSISM 1) includes all four indicators, and the second (NARCISSISM 2) includes the prominence of the CEO’s photograph in the company’s annual report and the CEO’s use of first-person singular pronouns in interviews. In our analysis, we will continue with NARCISSISM 2 as a measure for CEO narcissism.
3.2.3 Mediating variable – Audit Committee Status
For the mediating variable, audit committee status (STATUS), we will follow the research from Badolato et al. (2014), since they look at audit committee status as determinant of earnings management. In their measure of audit committee status, Badolato et al. (2014) include the number of: (1) contemporaneous public board directorships; (2) contemporaneous private board directorships; and (3) degrees. For these three variables applies that they will be measured as one if the mean of the number of directorships or degrees of the audit committee members is higher than the median of this number of all audit committees, and zero otherwise. This data will be obtained from BoardEx. Here, we will only look at the status of the audit committee itself, not at the relative status of the audit committee compared to the CEO, since the goal is to investigate
19
whether narcissistic CEOs affiliate with high status audit committee members and whether high status of audit committee members decreases the audit risk.
Two adjustments are made regarding the construct of Badolato et al. (2014). First, Badolato et al. (2014) calculated audit committee status as one if the sum of all three variables equals three, and zero otherwise. This means that audit committees either have high status or low status, and nothing in between. In her discussion of the article of Badolato et al. (2014), Hayes (2014) sees this as a weakness and, therefore, we make an adjustment in the measure. Thus, in this research, the sum of all three variables is equal to zero, one, two or three, which results in an audit committee status that can either be low, moderately low, moderately high, or high, respectively. For instance, when the mean of the contemporaneous public and private board directorships of an audit committee is higher than the median for all audit committees, and the mean of number of degrees in an audit committee is lower of the median for all audit committees, the sum of the indicator variables equals two. In the research of Badolato et al. (2014), this means that that the audit committee would have a low status, while in this research the status would be indicated as moderately high. Second, in our additional analysis we add financial expertise to the measure of status, since Veltrop et al. (2017) argue that financial expertise is an important indicator of status. We measure this as one if the audit committee has at least one member with financial expertise, and zero otherwise. Thus, the sum of all variables can then be equal to zero, one, two, three or four, which results in an audit committee status that can either be low, moderately low, neutral, moderately high, or high, respectively. We add this to the additional analysis, since we expect that the majority of the audit committees have at least one member with financial expertise. Therefore, the results of the main analysis could be biased, and we control for this effect in the additional analysis.
3.2.4 Moderating Variable – Audit Committee Size
There have been used several ways to measure the size of audit committees (ACSIZE). For example, Abbott et al., 2004 use a benchmark of minimum three audit committee member to be effective as audit committee, and therefore, code it as one if the audit committee had three or more members, and zero otherwise. Furthermore, Badolato et al. (2014) use the total number of audit committee members.
20
In this research, we will follow the direct measure of Badolato et al. (2014). Thus, ACSIZE is measured as the total number of audit committee members. This data will be collected from BoardEx.
3.2.5 Control variables
In the regression model, we also included control variables (CONTROLS). We control for these variables because, in existing literature, these variables have been associated with the dependent variable, the audit risk perceived by the auditor. First of all, we control for some firm-specific attributes. Johnstone (2000) discusses that auditors take the riskiness of the industry into account when assessing the audit risk before accepting the client. Therefore, we control for the industry-fixed effects, country-fixed effects and year-fixed effects for the firms. Furthermore, Krishnan and Visvanathan (2009) argue that firm specific variables like firm size (SIZE), measured as total assets in millions of euros; market value of equity (LMV), measured as the log of market value of equity; and return on assets before extraordinary items (ROA) can influence the audit risk. Therefore, we control for these variables, just as for the leverage (LEV), measured as total liabilities over total assets; and for loss-making firms (NROA), measured as 1 if firms have a negative ROA, and 0 otherwise. Furthermore, we control for the inherent risk (IHRISK) of the firm, measured as total inventories and receivables over total assets, like Hay et al. (2008) did.
Second, auditor-related variables are taken into account, since the auditor assesses the audit risk. Therefore, we control for audit company-fixed effects. Also, we will control for audit fees (AUDFEES%), measured as the total annual audit fees over total assets. The data for auditor size and change will be hand-collected, and the data for audit fees will be collected from Datastream.
Third, we control for CEO related variables, since these characteristics could also influence the proposed relationships. We control whether the CEO characteristics age (CEOAGE), tenure (CEOTEN), and gender (CEOGEN) have influence on the proposed relationships. CEOAGE and
CEOTEN are measured in years, and CEOGEN is measured as 1 if the CEO is a male, and 0 if the
CEO is a female. The CEO data will be collected from Datastream.
Fourth, board and audit committee variables are controlled for, since these characteristics can influence the audit risk of a company (Abbott et al., 2003; Abbott et al., 2004). Therefore, we control for board size (BSIZE), measured as total number of board members; board independence (BIND), measured as a percentage of independent board members of the total number of board
21
members; and board meetings (BMEET), measured as the total number of annual board meetings. Furthermore, we control for audit committee independence (ACIND), measured as the percentage of independent audit committee members of the total number of audit committee members. This data will be obtained from Datastream and BoardEx.
Finally, we control for two ownership variables, available on Datastream. Krishnan and Visvanathan (2009) argue that a high degree of inside ownership (INOWN) decreases the need for an extensive audit, because these firms face lower agency costs. INOWN, according to Krishnan and Visvanathan (2009), is measured as the cumulative percentage of voting control held by managers and directors. Furthermore, the authors state that block shareholders (BLOCK), measured as the cumulative ownership percentage of voting control held by block holders, is positively related to substantive auditing procedures. Since block holders own at least five percent of voting control, they have a higher demand for assurance to monitor the organization (Krishnan & Visvanathan, 2009).
22 4. RESULTS
4.1 Descriptive Statistics
Table 2 provides the descriptive statistics of our variables. This table shows us the number of observations, the mean, median, standard deviation, minimum and maximum of the variables in the dataset. The variables are winsorized, to prevent that outliers in the data can influence the results. Winsorizing means that outliers in the data are replaced, instead of omitted. The values of the below the first and above the 99th percentile of the variables are replaced by the value of the
first and 99th percentile. Also, we used the natural logarithm of the total assets (LnSIZE), market value (LMV) and for materiality (LnMAT). Audit fees (AUDFEES%) are calculated as a percentage of the firm’s total assets. Furthermore, due to the limited availability of the CEO compensation data, we used the NARCISSISM 2 measure, instead of NARCISSISM 1, to increase or dataset. Finally, also in order to enlarge the dataset, we replaced the missing values of the control variables by the mean of that specific variable, or the median in case of dummy variables.
Variables n Mean Median Std. Dev. Min Max
RISK 849 3,7326 4,0000 1,3266 1,0000 7,0000 NARCISSISM 2 849 0,0424 0,0190 0,4217 -0,6829 1,4723 STATUS 849 1,2945 1,0000 0,9703 0,0000 3,0000 ACSIZE 849 3,5524 3,0000 0,9541 2,0000 7,0000 LnSIZE 849 14,1615 14,0791 1,7917 10,2133 19,1825 LMV 849 7,2681 0,7280 1,8871 2,3263 11,6664 ROA 849 0,0615 0,0600 0,0920 -0,3552 0,3049 LEV 849 0,5498 0,5572 0,2127 0,0706 1,1433 NROA 849 0,1296 0,0000 0,3360 0,0000 1,0000 IHRISK 849 0,2770 0,2463 0,1956 0,0108 0,9206 AUDFEES% 849 0,0085 0,0010 0,0541 0,0001 0,4645 CEOAGE 849 52,7257 52,3000 5,4387 39,0000 70,0000 CEOTEN 849 6,4515 5,5000 5,0517 0,2000 27,0000 CEOGEN 849 0,9741 1,0000 0,1590 0,0000 1,0000 BSIZE 849 9,0059 9,0000 1,8515 5,0000 15,0000 BIND 849 0,4999 0,5045 0,1603 0,0558 0,8341 BMEET 849 8,7079 9,0000 2,1696 4,0000 18,0000 ACIND 849 0,9299 1,0000 0,1299 0,3333 1,0000 LnMAT 849 8,5272 8,4007 1,7147 5,0106 13,3047 INOWN 849 0,0707 0,0000 0,1396 0,0000 0,5900 BLOCK 849 0,0824 0,0600 0,0901 0,0000 0,4200
23
We can see that there are 849 firm-year observations. In this research, the initial sample consisted of firms in the UK and the Netherlands. However, due to de limited data availability for Dutch firms, we continue with the UK sample. Although the Dutch firms are excluded in the main analysis, it is still possible to control for country fixed effects, since the sample there are firms in the UK sample which are incorporated in different countries. Also, the Dutch firms will be included in the additional analysis. Furthermore, we also control for the industry-fixed effects, year effects and auditor-fixed effects. The specification of the countries and the industries can be found in table 3. The firm observations per year and audit company are summarized in table 4. Here we can see that the majority of the firm observations are incorporated in Great Britain and operate in the industrials industry. Furthermore, EY is quite underrepresented compared to the other Big 4 firms, as auditor of the sample firms.
Basic materials
Consumer goods
Consumer
services Healthcare Industrials Oil & Gas
Techno-logy
Telecomm-unications Utilities Total
BM 3 0 0 0 0 0 0 0 0 3 CA 0 0 3 0 0 0 0 0 0 3 CH 0 1 0 0 0 0 0 0 0 1 GB 50 104 155 48 280 34 38 20 8 737 GG 0 0 0 0 3 0 0 0 0 3 GI 0 0 4 0 0 0 0 0 0 4 IE 0 14 10 0 15 4 0 0 0 43 IM 0 0 4 0 0 3 0 0 0 7 JE 14 0 7 4 3 8 0 0 4 40 SG 0 0 0 0 2 0 0 0 0 2 US 0 0 3 0 0 0 0 0 0 3 VG 3 0 0 0 0 0 0 0 0 3 Total 70 119 186 52 303 49 38 20 12 849
Table 3 - Country and Industry summary
Industry C o u n tr y o f I n c o r p o r a ti o n
Deloitte EY KPMG PwC Other Total
2013 37 19 45 39 9 149 2014 59 31 72 68 16 246 2015 58 35 72 61 15 241 2016 51 32 61 56 13 213 Total 205 117 250 224 53 849 Ye a r Audit company
4.2 Correlation Matrix
The table above presents the correlation matrix of the variables. With the correlation matrix it is possible to assess whether there is a case of multicollinearity. Multicollinearity means that the regression model can be predicted with one of the explanatory variables, which correlates with another explanatory variable. Multicollinearity occurs when the correlation is higher than 0,7 or lower than -0,7. As we can see, the narcissism measures correlate with each other. This can be explained, since the measures of NARCISSISM 1 and
NARCISSISM 2 have overlap. Furthermore, firm size (SIZE) and log of the market value of equity (LMV) correlate, because they are Variables RIS K N A R C ISSI SM 2 S T AT US AC S IZ E L n S IZ E L M V R OA L E V NR OA IH R IS K AUDF E E S % C E OAGE C E O T E N C E OGE N B S IZ E B IND B M E E T AC IND L n M AT INOW N BL O CK RISK 1 NARCISSISM 2 0,1939 1 STATUS 0,0832 0,1447 1 ACSIZE 0,1299 0,1645 0,0929 1 LnSIZE 0,3870 0,2499 0,3392 0,3888 1 LMV 0,2819 0,2025 0,2977 0,4363 0,8873 1 ROA -0,1285 -0,0958 -0,1236 0,1596 0,0212 0,2350 1 LEV 0,1874 0,0737 -0,0082 0,1207 0,2107 0,0786 0,0007 1 NROA 0,1122 0,0659 0,0782 -0,1279 -0,0577 -0,1668 -0,6608 -0,0543 1 IHRISK -0,1607 -0,1808 -0,1336 -0,0109 -0,2484 -0,2241 0,1674 0,0993 -0,1343 1 AUDFEES% -0,1118 -0,1705 -0,1475 -0,1127 -0,2560 -0,3092 0,0086 -0,0143 -0,0464 0,2339 1 CEOAGE -0,0394 -0,0412 -0,0460 0,0199 -0,0045 0,0011 0,0036 -0,0840 0,0138 0,0540 0,0167 1 CEOTEN -0,1748 -0,2878 -0,1749 -0,0165 -0,1813 -0,1492 0,1010 -0,0889 -0,0505 0,2128 0,0776 0,2746 1 CEOGEN -0,0161 -0,0790 0,0266 -0,0532 -0,0726 -0,0855 0,0024 0,0030 0,0188 0,0905 -0,0407 0,0377 0,0315 1 BSIZE 0,2484 0,0986 0,1507 0,3473 0,3930 0,4052 0,0262 0,0973 -0,0335 -0,1387 0,0009 0,0365 -0,0459 -0,0916 1 BIND 0,1373 0,0138 0,0181 0,1764 0,2164 0,1908 0,0302 0,0336 0,0025 -0,0820 0,0111 0,0729 -0,1039 0,0215 0,1678 1 BMEET 0,0322 0,0333 -0,0403 -0,0422 -0,0817 -0,1019 -0,0233 0,0757 0,0164 -0,0299 0,0339 -0,0520 -0,0562 0,0464 -0,0398 -0,0193 1 ACIND 0,0311 -0,0014 -0,0058 0,0539 0,0834 0,0700 -0,0403 0,0362 0,0580 0,0184 0,0053 0,0348 -0,0918 0,0044 0,0798 0,4662 -0,0681 1 LnMAT 0,3274 0,2030 0,3263 0,4011 0,9298 0,8994 0,1226 0,1240 -0,0527 -0,2089 -0,2339 0,0384 -0,1438 -0,0567 0,4196 0,2244 -0,1185 0,0958 1 INOWN -0,0905 -0,0928 -0,1041 -0,2148 -0,1903 -0,2179 -0,0046 -0,0512 0,1187 0,0576 0,1811 -0,0284 0,1402 -0,0018 -0,0532 -0,1004 -0,0676 -0,0684 -0,1308 1 BLOCK -0,0988 -0,0672 0,0162 -0,0473 -0,1179 -0,1228 -0,0526 0,0424 0,0406 0,0358 -0,0350 -0,0101 0,0691 -0,0121 -0,1139 -0,0063 0,0087 -0,0281 -0,1593 -0,1304 1 Table 5 - Correlation Matrix
both measures for firm size. Materiality (MATERIALITY) correlates with both firm size and log of the market value of equity. This seems logical, since the materiality is often assessed by using a percentage of a certain benchmark, which can be, for instance, total assets. Since we want to rule out multicollinearity, we exclude the variables with the lowest correlation to audit risk (RISK), which are the LMV and LnMAT.
Furthermore, we see a positive correlation between NARCISSISM 2 and RISK, which could indicate that there is a positive relationship between these variables. This will be tested in the following sections. Another noteworthy point of attention is the negative correlation between
IHRISK and RISK, since one would expect that the audit risk will be higher when the inherent risk
is high. This correlation suggests the opposite.
4.3 Main analysis
Table 6 presents the regression results for both hypotheses. The coefficients are the direct effects, of which the indirect and total effects can be derived. As discussed, for our first hypothesis, we test whether audit committee status mediates the relationship between CEO narcissism and audit risk (Model 1 and Model 2). Expected is that CEO narcissism leads to an increase in status of audit committee members, resulting in either a low or a high number of audit risks. First of all, the results point out that there is a positive significant relationship between CEO narcissism and audit risk (β= 0.174, p= 0.068). Furthermore, the results show that there is a no significant relationship between CEO narcissism and audit committee status (β= -0.0262, p= 0.739) and a negative and non-significant relationship between audit committee status and audit risk (β= -0.0513, p= 0.216). When we combine this, we come to the total effect of this mediating relationship. This can be calculated by multiplying the coefficient of the relationship between CEO narcissism and audit committee status with the coefficient of the relationship between audit committee status and audit risk, and adding this to the coefficient of the direct relationship between CEO narcissism and audit risk. This results in a total effect of β= 0.175, which points out to be significant at the p<0.1 level (p= 0.066). This could indicate that audit committee status mediates the relationship between CEO narcissism and audit risk. However, the indirect effect of the effect in this mediating relationship is very small with a delta in the coefficient of β= 0.001, and a direct effect that is still significant.
26
Model (1) Model (2) Model (3) Model (4)
VARIABLES STATUS RISK STATUS RISK
STATUS -0.0513 -0.315** (0.0415) (0.155) NARCISSISM 2 -0.0262 0.174* -0.0262 0.192** (0.0787) (0.0951) (0.0787) (0.0956) ACSIZE -0.137** (0.0691) ModSIZE 0.0733* (0.0418) LnSIZE 0.164*** 0.254*** 0.164*** 0.257*** (0.0220) (0.0274) (0.0220) (0.0281) ROA -1.025** -1.300** -1.025** -1.309** (0.447) (0.543) (0.447) (0.548) LEV -0.349** 0.422** -0.349** 0.423** (0.158) (0.192) (0.158) (0.192) NROA 0.0209 0.333** 0.0209 0.333** (0.122) (0.148) (0.122) (0.148) IHRISK 0.0807 0.200 0.0807 0.230 (0.189) (0.229) (0.189) (0.229) AUDFEES% -0.801 -1.111 -0.801 -1.244 (0.635) (0.769) (0.635) (0.770) CEOAGE -0.00617 0.00359 -0.00617 0.00400 (0.00605) (0.00732) (0.00605) (0.00732) CEOTEN -0.0226*** -0.0105 -0.0226*** -0.0106 (0.00681) (0.00829) (0.00681) (0.00831) CEOGEN 0.239 0.399* 0.239 0.362 (0.192) (0.232) (0.192) (0.232) BSIZE 0.0603*** 0.0539** 0.0603*** 0.0568** (0.0186) (0.0226) (0.0186) (0.0231) BIND -0.440** 0.333 -0.440** 0.367 (0.220) (0.266) (0.220) (0.267) BMEET -0.00238 0.0168 -0.00238 0.0161 (0.0142) (0.0172) (0.0142) (0.0171) ACIND -0.385 -0.430 -0.385 -0.441 (0.273) (0.330) (0.273) (0.330) INOWN -0.316 -0.444 -0.316 -0.509* (0.239) (0.290) (0.239) (0.292) BLOCK 0.996*** -0.562 0.996*** -0.575 (0.356) (0.433) (0.356) (0.432) Constant -1.607** -0.860 -1.607** -0.460 (0.745) (0.904) (0.745) (0.926) Observations 849 849 849 849
Year-fixed effects Yes Yes Yes Yes
Industry-fixed effects Yes Yes Yes Yes
Country-fixed effects Yes Yes Yes Yes
Audit company-fixed effects Yes Yes Yes Yes
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Hypothesis 1 Hypothesis 2
27
Furthermore, the relationship between CEO narcissism and audit committee status and the relationship between audit committee status and audit risk are not significant. Therefore, we are not able to accept H1, and cannot conclude that audit committee status mediates the relationship between CEO narcissism and audit risk.
The results of the second hypothesis are presented in Model 3 and Model 4 of table 6. As we can see, Model 1 and Model 3 are similar. The difference is that we test whether the size of an audit committee moderates the proposed mediating relationship, and that does not influence the relationship on audit committee status. We hypothesized that audit committee size strengthens the mediating role of audit committee status in the relationship between CEO narcissism and audit risk. The results point out that the relationship between audit committee status and audit risk is stronger when audit committee size is added as moderator, and significant at the p<0.5 level (β= -0.315, p= 0,043). Also, the interaction variable (ModSIZE) is significant at the p<0.1 level (β= 0.0733, p= 0.080). Furthermore, the results show us that the direct effect of CEO narcissism is stronger, and even more significant (β= 0.192, p= 0.045) and that the total effect is still positive and significant (β= 0.200, p= 0.043). Although the interaction variable is significant and it strengthens the relationship between audit committee status and audit risk, we are also not able to conclude that audit committee size strengthens the mediating role of audit committee status in the relationship between CEO narcissism and audit risk, since this mediating effect does not exist. Therefore, we cannot accept H2.
In our additional analysis we will perform a similar structural equation method, with different measures of variables and with the Dutch sample firms included. Furthermore we will test whether there exist significant relationships when we use the linear regression method.
4.4 Additional analysis
4.4.1 Structural equation model
In this additional analysis the same structural equation model will be performed, however, the variables are slightly different. First of all, we use the number of abnormal audit risks (ABRISK) instead of all reported key audit matters. We calculate this as the number of audit risks minus 1, the minimum number of audit risks in the sample. Second, as discussed in the methodology section, we add financial expertise as an indicator variable to the status measure of Badolato et al. (2014). This results in the new variable ACSTATUS. Finally, we include the Dutch sample firms in the total sample.
28
Table 7 provides the results of the additional analysis. The results in Model (1) and Model (2) point out that there is a significant positive relationship between CEO narcissism and abnormal audit risks at the p<0.05 level (β= 0.199, p= 0.031). Furthermore, there is no significant relationship between CEO narcissism and audit committee status (β= 0.0076, p= 0.919). Similar to the main analysis, the results show a non-significant negative relationship between audit committee status and audit risk (β= -0.0419, p= 0.286). The total effect of the mediating relationship points out to be positive, and significant at the p<0.05 level (β= 0.198, p= 0.032). This would indicate that a mediating relationship exists. However, similar to the main analysis, the relationships between CEO narcissism and audit committee status and the relationship between audit committee status and audit risks are non-significant. Furthermore, the indirect effect of the possible mediation is very small with a delta of β= -0.001. Therefore, we are not inclined to withdraw our conclusion in our main analysis, where we did not accept H1 and concluded that there was no mediating relationship
For the second hypothesis, we used the same variables as in the additional analysis for hypothesis 1 and added the moderator to the model. The results are shown in Model (3) and Model (4) in table 7. Model (3) is once again similar to Model (1), just as in the main analysis. Model (4) shows that there is a significant positive relationship between CEO narcissism and audit risk (β= 0.220, p= 0.018), and an significant negative relationship between audit committee status and audit risk (β= -0.279, p= 0.059). Thus, we see that when the moderator is added in the model, the coefficient further decreases. Also, the interaction variable (ModSIZE) is significant (β= 0.0663, p= 0.096). Hence, we can conclude that audit committee size strengthens the relationship between audit committee status and audit risk. However, since there is no mediating relationship, we still cannot accept hypothesis 2.
Furthermore, we see that the total effect of the relationship between CEO narcissism and audit risk, with audit committee status as a mediator is positive and significant (β= 0.218, p= 0.022). Thus, the results of this additional analysis give us similar results as in the main analysis. Both analyses show a significant total effect in the mediating relationship, however, the indirect effects are small and insignificant. Furthermore, both analyses show a non-significant relationship between audit committee status and audit risk. However, when there is a larger audit committee, this effect will be significantly stronger.
29
Model (1) Model (2) Model (3) Model (4)
VARIABLES ACSTATUS ABRISK ACSTATUS ABRISK
ACSTATUS -0.0419 -0.279* (0.0393) (0.148) NARCISSISM 2 0.00760 0.199** 0.00760 0.220** (0.0748) (0.0925) (0.0748) (0.0932) ACSIZE -0.192* (0.0989) ModSIZE 0.0663* (0.0399) LnSIZE 0.155*** 0.228*** 0.155*** 0.232*** (0.0202) (0.0257) (0.0202) (0.0264) ROA -1.060** -1.226** -1.060** -1.236** (0.422) (0.523) (0.422) (0.528) LEV -0.361** 0.447** -0.361** 0.451** (0.149) (0.185) (0.149) (0.184) NROA -0.0135 0.353** -0.0135 0.353** (0.113) (0.140) (0.113) (0.140) IHRISK 0.0417 0.173 0.0417 0.199 (0.179) (0.221) (0.179) (0.221) AUDFEES% -0.920 -1.150 -0.920 -1.262* (0.580) (0.718) (0.580) (0.719) CEOAGE -0.00999* 0.00177 -0.00999* 0.00201 (0.00564) (0.00698) (0.00564) (0.00698) CEOTEN -0.0182*** -0.00665 -0.0182*** -0.00615 (0.00635) (0.00788) (0.00635) (0.00791) CEOGEN 0.210 0.276 0.210 0.245 (0.177) (0.219) (0.177) (0.219) BSIZE 0.0564*** 0.0617*** 0.0564*** 0.0649*** (0.0172) (0.0213) (0.0172) (0.0218) BIND -0.276 0.351 -0.276 0.381* (0.185) (0.230) (0.185) (0.230) BMEET 0.00486 0.0100 0.00486 0.00981 (0.0130) (0.0161) (0.0130) (0.0160) ACIND -0.118 -0.160 -0.118 -0.174 (0.185) (0.229) (0.185) (0.229) INOWN -0.196 -0.400 -0.196 -0.465* (0.224) (0.278) (0.224) (0.280) BLOCK 0.875*** -0.872** 0.875*** -0.897** (0.329) (0.408) (0.329) (0.408) Constant -0.348 -0.648 -0.348 -0.0663 (0.604) (0.747) (0.604) (0.820) Observations 991 991 991 991
Year-fixed effects Yes Yes Yes Yes
Industry-fixed effects Yes Yes Yes Yes
Country-fixed effects Yes Yes Yes Yes
Audit company-fixed effects Yes Yes Yes Yes
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Table 7 - Additional analysis: regression results from structural equation modeling
30
4.4.2 Linear regression analysis
In this additional analysis, separate linear regressions will be executed, in order to find out whether the outcomes are similar to those from the structural equation model in the main tests.
The analysis for the first hypothesis consists of four steps and are summarized in Model (1) until Model (4) in table 8, respectively. First, we conduct a regression analysis where CEO narcissism predicts audit risk. Second, we conduct a regression analysis where CEO narcissism predicts audit committee status. Third, a regression analysis is conducted with audit committee status predicting audit risk. Finally, we conduct a multiple regression analysis where both CEO narcissism and audit committee status predict audit risk. The second hypothesis consists of two multiple regressions, summarized in Model (5) and Model (6) in table 8. The first regression for hypothesis 2 consists of audit committee status and the moderator audit committee size as predictors for the audit risk. The second regression calculates the total effect of the moderated mediating relationship, where CEO narcissism and audit committee status predict audit risk, moderated by the audit committee size.
The results in Model (1) until Model (4) in table 8 show that the direct relationship between CEO narcissism and audit risk is positive and significant at the p<0.10 level (β= 0.175, p= 0.074); the relationship between CEO narcissism and audit committee status is negative, non-significant (β= -0.0262, p= 0.745); the relationship between audit committee status and audit risk is negative, but not significant (β= -0.0522, p= 0.221); and finally, the total effect of the relationship, where we hypothesized that audit committee status mediates between CEO narcissism and audit risk, is significant (β= 0.174, p= 0.076). Thus, we can conclude that the regression analysis provides similar results compared to the structural equation model in the main analysis. Thus, we still cannot accept hypothesis 1, since the indirect effect (i.e., the relationship between CEO narcissism and audit committee status and the relationship between audit committee status and audit risk) is not significant, and therefore, no mediating relationship exists between CEO narcissism and audit risk. Model (5) and Model (6) in table 8 show the results for the second hypothesis. Here also applies that the results are quite similar to those in the structural equation model in the main analysis. Model (5) shows that there is a non-significant negative relationship between audit committee status and audit risk (β= -0.302, p= 0.059). However, in this model it is not significantly strengthened by the size of an audit committee (β= 0.0697, p= 0.106).
31
Model (1) Model (2) Model (3) Model (4) Model (5) Model (6)
VARIABLES RISK STATUS RISK RISK RISK RISK
NARCISSISM 2 0.175* -0.0262 0.174* 0.192* (0.0977) (0.0807) (0.0977) (0.0983) STATUS -0.0522 -0.0513 -0.302* -0.315** (0.0427) (0.0426) (0.160) (0.160) ACSIZE -0.122* -0.137* (0.0708) (0.0711) ModSIZE 0.0697 0.0733* (0.0430) (0.0430) LnSIZE 0.245*** 0.164*** 0.258*** 0.254*** 0.261*** 0.257*** (0.0273) (0.0226) (0.0281) (0.0282) (0.0289) (0.0289) ROA -1.247** -1.025** -1.351** -1.300** -1.377** -1.309** (0.556) (0.459) (0.557) (0.557) (0.563) (0.563) LEV 0.440** -0.349** 0.442** 0.422** 0.445** 0.423** (0.197) (0.163) (0.197) (0.197) (0.197) (0.197) NROA 0.332** 0.0209 0.334** 0.333** 0.333** 0.333** (0.152) (0.126) (0.152) (0.152) (0.152) (0.152) IHRISK 0.196 0.0807 0.188 0.200 0.214 0.230 (0.235) (0.194) (0.235) (0.235) (0.236) (0.235) AUDFEES% -1.070 -0.801 -1.174 -1.111 -1.302 -1.244 (0.789) (0.652) (0.790) (0.790) (0.793) (0.792) CEOAGE 0.00391 -0.00617 0.00423 0.00359 0.00471 0.00400 (0.00752) (0.00621) (0.00752) (0.00752) (0.00753) (0.00752) CEOTEN -0.00930 -0.0226*** -0.0139* -0.0105 -0.0144* -0.0106 (0.00846) (0.00699) (0.00830) (0.00851) (0.00833) (0.00855) CEOGEN 0.387 0.239 0.373 0.399* 0.336 0.362 (0.238) (0.197) (0.238) (0.238) (0.239) (0.239) BSIZE 0.0509** 0.0603*** 0.0558** 0.0539** 0.0578** 0.0568** (0.0231) (0.0191) (0.0233) (0.0232) (0.0238) (0.0238) BIND 0.356 -0.440* 0.307 0.333 0.332 0.367 (0.273) (0.226) (0.274) (0.274) (0.274) (0.274) BMEET 0.0169 -0.00238 0.0185 0.0168 0.0179 0.0161 (0.0176) (0.0146) (0.0176) (0.0176) (0.0176) (0.0176) ACIND -0.410 -0.385 -0.446 -0.430 -0.456 -0.441 (0.339) (0.280) (0.340) (0.339) (0.339) (0.339) INOWN -0.428 -0.316 -0.470 -0.444 -0.527* -0.509* (0.297) (0.245) (0.297) (0.297) (0.301) (0.300) BLOCK -0.613 0.996*** -0.572 -0.562 -0.584 -0.575 (0.443) (0.366) (0.445) (0.445) (0.445) (0.444) Constant -0.778 -1.607** -0.965 -0.860 -0.600 -0.460 (0.926) (0.765) (0.928) (0.928) (0.952) (0.953) Observations 849 849 849 849 849 849
Year-fixed effects Yes Yes Yes Yes Yes Yes
Industry-fixed effects Yes Yes Yes Yes Yes Yes
Country-fixed effects Yes Yes Yes Yes Yes Yes
Audit company-fixed effects Yes Yes Yes Yes Yes Yes
R-squared 0.405 0.241 0.404 0.406 0.406 0.408
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Hypothesis 1 Hypothesis 2
32
Furthermore, Model (6) shows the total effect of the moderated mediating relationship. The results here point out that CEO narcissism is positively related to audit risk (β= 0.192, p= 0.052). Furthermore, the interaction variable ModSIZE is also significant (β= 0.0733, p= 0.089). Therefore, we see that the results are robust, since they are similar to those in the structural equation method in the main analysis.
