MSc Accountancy & Control, track Control
Master thesis:
Does one size fit all? Evidence from board gender
diversity
Faculty of Economics and Business, University of Amsterdam
Student: Jelle Tessel Student number: 10572856 Supervisor: Dr. B. Qin Date: 19th June 2015 Word count: 12,954
2
Statement of Originality
This document is written by student Jelle Tessel who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
In this paper I test whether one size fits all when it comes to board gender diversity. I am interested whether the uniformity of corporate governance structures is also the case for board gender diversity. I use the data from the databases “ISS” and “Compustat” for the U.S. public-listed companies of the S&P 1500 in the years 2006 till 2010. The sample consists of 365 unique firms yielding 2,190 firm years. I collect financial data such as equity, debt, leverage, net sales and assets and corporate governance data such as the gender of the board members, the independence status and the number of seats of the board.
After doing principal component factor analyses, the results of the regressions reveal that less complex companies are more likely to appoint female directors in their boards. Furthermore, a positive relationship exists between female directors and the riskiness of the company. This can be explained by the glass cliff theory that argues that risky companies are more likely to appoint female directors because female directors take more responsibility and are better able to manage people under stressful circumstances. I find no evidence that a relationship exists between board gender diversity and loss-making companies. Regarding the mismatch in my empirical model that I use to predict board gender diversity, the subsequent financial performance is not considerable different for companies that deviate from this empirical model as for the companies that do not deviate. Based on these results it can be said that the one size fits all assumption for board gender diversity does not hold for some firm characteristics, such as riskiness of the company. In other words, considering the firm’s distinct risk profile, it is not always the case that the greater board gender diversity the better. My study contributes to the prior literature of Faleye (2007) that also focus on the uniformity of corporate governance structures but than in the case of CEO duality instead of board gender diversity. Besides, my study also supports the board gender diversity studies that focus on the relationship between board gender diversity and financial performance. Instead of the study of Carter, D’Souza, Simpkins and Simpson (2010), my study does show that board gender diversity has a positive relationship with the financial performance of a company like the study of Bilimoria (2006). My study also supports the glass cliff theory developed by Ryan and Haslam in 2004 that explains why risky companies are more likely to appoint female directors in their boards.
4
Table of contents
1 Introduction ... 5
2 Literature review and hypotheses ... 7
2.1 Contingency theory ... 7
2.2 Theories related to gender diversity ... 8
2.2.1 The ceiling theory ... 8
2.2.2 Token theory ... 9
2.3 Explanations for board gender diversity ... 10
2.3.1 Person-centered explanations ... 10
2.3.2 Situation-centered explanations ... 11
2.3.3 Social system-centered explanations ... 11
2.4 Hypothesis development ... 12
3 Methodology ... 16
3.1 Measurement of board gender diversity ... 16
3.2 Measures for operational complexity... 16
3.3 Measures for risk taking behavior ... 16
3.4 Measure for loss-making companies ... 17
3.5 Measures for subsequent financial performance ... 17
3.6 Control variables ... 18
3.7 Empirical models ... 18
4 Data and models ... 21
4.1 Sample ... 21
4.2 Descriptive statistics ... 21
4.3 The correlationmatrix ... 26
4.4 The relationship between board gender diversity and firm characteristics... 28
4.5 How to tackle the potential correlation among the variables in the regressions? ... 31
4.6 The relationship between the mismatch in board gender diversity and subsequent financial performance ... 34
4.7 The robustness tests of the residual ... 37
5 Summary and conclusion ... 41
5 1 Introduction
According to Hermalin and Weisbach (1991) does one size fit all not exists for corporate governance structures. The authors argue that every company is unique and has its own identical corporate governance characteristics. This opinion is supported by the ABA (2004) that argues that there is no “one” corporate governance structure. Every company should develop its own corporate governance structure that fits best to the identical nature and circumstances of the company. One study against the uniformity of corporate governance structures is the study of Yu (2005). The author argues that an explanation for not supporting the one size fits all assumption for corporate governance structures is because of the diversity of corporate financing methods that can cause a conflict of interest. The solutions to solve these conflicts of interest are diverse and can therefore not be solved by “one” corporate governance structure (Yu, 2005).
However, the Task Force supports the one size fits all assumption because it argues that the uniformity of governance structures will lead to an increase in the responsibility of the management of the company (Culp and Niskanen, 2003). Gompers, Ishii and Metrick (2003) argue that the usage of one single corporate governance structure has to be promoted because this could increase the performance of the firm.
Gender diversity in the board of directors is a contemporary phenomenon that is used in a lot of academic studies. In addition, the society and government are also concerned with gender diversity in the board of directors. A good example is the question whether or not mandating female quotas for the appointment of female directors in the board. These female quotas can be seen as an attempt to uniform the gender diversity in the board of directors in the different business industries. The study of Ahern and Dittmar (2012) is dedicated to this phenomenon in Norwegian where companies must have at least 40 percent of female directors in their boards as a result of a law in 2003. The mean massage of the study is that maximum firm value is achieved when companies can voluntary choose the gender composition of their boards. Ahern and Dittmar (2012) argue that the female quota leads to unexperienced and younger boards that increase the leverage and decrease firm performance.
My motivation for this study is to discover whether uniformity of corporate
governance structures exists when it comes to gender diversity. I would like to examine if it is possible to predict board gender diversity by the usage of a statistical model. I am especially interested if the uniformity of board gender diversity exists in the audit, compensation, nomination and finance committees of U.S. public-listed firms. For these committees, a lot of
6 legislation is already applicable that causes uniformity of corporate governance structures in those committees such as SOX. Because of this uniformity in these committees due to
legislation, I am interested if a voluntary uniformity of corporate governance structures exists when it comes to board gender diversity.
To find this out, I do a comparable study as the study of Faleye (2007). This study of Faleye (2007) is also related to the one size fits all assumption but in the context of CEO duality instead of board gender diversity. Although this study of Faleye (2007) is executed in another context it gives me guidance for the structure of the research methodology.
My study is interesting in a societal point of view because I deviate from previous studies related to board gender diversity. Prior studies are mainly focused on the relationship between board gender diversity and firm performance. These studies are especially interested in finding the optimal corporate governance structure that creates maximum firm value. My study on the other hand is not only related to the relationship between board gender diversity and financial performance but is also examining if companies extend their uniformity of corporate governance structures when it comes to board gender diversity. If this uniformity in board gender diversity exists, I would like to find out if this can be explained by the financial and/or governance characteristics of the company.
I use the data related to the companies of the S&P 1500 of the years 2006 till 2010. The sample for this study consists of 365 unique firms yielding 2,190 firm years. After doing principal component factor analyses, the regressions show that the one size fits all assumption for board gender diversity is applicable on some firm characteristics. I find evidence that less complex companies are more likely to appoint female directors. Besides, there exists a positive relationship between the riskiness of the company and the appointment of female directors that can be explained by the glass cliff theory developed by Ryan and Haslam in 2004. I find no evidence for a negative relationship between loss-making companies and board gender diversity. Furthermore, the companies that deviate (mismatch) in my empirical model that I use for predicting board gender diversity have almost the same relationship with subsequent financial performance as the companies that do not deviate.
The following parts of this study consists of the literature review and the hypotheses (part 2), the description of the variables in combination with my empirical models (part 3), the sample, regressions and results of the regressions (part 4) and in the last part (part 5) I give the summary, conclusions, limitations and recommendations for future research.
7 2 Literature review and hypotheses
2.1 Contingency theory
The contingency theory is about how the leadership style of a manager fits to the job (Lussier and Achua, 2014). According to Fiedler (1967) depends the effectiveness of a manager’s leadership style to the concept of fit. Fiedler (1967) argues that the concept of fit depends not on changing a manager’s leadership style to the job but on how well the manager change the context of the job in such manner that it fits the best to his/her leadership style (Fiedler, 1967). Maximum firm performance is achieved when the situation fits the most within the manager’s leadership style (Fiedler, 1967). Van de Ven and Drazin (1985) describe fit as when a match exists between two factors or more.
Lussier and Achua (2014) argue that the contingency theory of leadership style developed by Fiedler is nowadays also used in other contexts. The contingency theory is for example used to explain what the determinants are for the effectiveness of the firm (Ghofar and Islam, 2015). Donaldson (2001) argues that a relationship exists between the structure, the environment and the strategy of a company. The environment and strategy of a company are the contingency factors because these factors are the determinants for creating firm value (Donaldson, 2001). Donaldson (2001) argues that the business environment affects the operations of the company and that the business environment is continuous changing. This continuous change of the business environment force companies to adjust their strategy on this environment to be successful.
Ghofar and Islam (2015) argue that the contingency theory can be used to explain the functioning of corporate governance. Corporate governance helps suppliers of capital for getting a return on their investment (Shleifer and Vishny, 1997). Besides, corporate governance is about rights and responsibilities and how these are distributed to the
companies’ participants such as the board of directors, the managers but also to shareholders and other stakeholders (European Central Bank, 2004). Ghofar and Islam (2015) argue that in the context of the contingency theory, the determinants for creating firm value are the same as for corporate governance. This means that corporate governance is also determined by the business environment and business strategy. Corporate governance consists of an internal and external mechanism. The internal mechanism is related to the monitoring of the executive board by the board of directors while the external mechanism is the market for corporate control (Daily, Dalton and Cannella, 2003). Ghofar and Islam (2015) argue that the business
8 competition is a characteristic of the business environment and this business competition leads to a decrease of internal governance problems (i.e. agency problems) since the company has to differentiate itself from competitors. The relationship between corporate governance and business strategy is in practice mostly related to the board characteristics (i.e. board size, independence, etc.) and the business strategy.
The contingency theory has also been criticized because the theory is not clear enough and does not specify enough the relationship between variables (Husted, 2000) but if the assumptions for these relationships are explicit described, these variables are suitable for statistical testing (Doty and Glick, 1994).
2.2 Theories related to gender diversity
2.2.1 The ceiling theory
The ceiling theory refers to the glass ceiling which hinder women to reach a position in the top management of a company (Oakley, 2000). Morrison, White, Van Velsor and The Center for Creative Leadership (1987) describe the glass ceiling as an invisible barrier that prevent women to reach a higher stair on the corporate ladder what means that women get stuck on a certain level. The glass ceiling makes a distinction between men and women not because of the differences in job and education qualifications but because of gender (Morrison and Von Glinow, 1990).
An explanation for the existence of the glass ceiling is because the top management prefers members who fit within the management team above the job qualifications of the members (Powell, 1999). Oakley (2000) argues that men are afraid that the status quo will change if women are appointed in the top management. This status quo is often called the old boy network and could change if a woman is appointed in the top management because men can lose their grip on the situation. Oakley (2000) describes the old boy network as an informal male social system in the top of a company that excludes all women and less
powerful men from having a top position in the company. This informal male system assigns formal positions based on informal relationships such as friendships (Oakley, 2000). The old boy network remains to exist by for example the usage of competency testing. This is a process in which women have to demonstrate that they are suited for a top management function (Oakley, 2000). The study of Rosener (1995) shows that competency testing is more often used for women than men when applying for top management functions. Consistent
9 with these findings is the statement of Powell (1999) who argues that the selection of top managers is more biased by decision makers than the selection for lower managers. Some countries are aware of this glass ceiling and would like to break through this informal system by using a female quota for female directors in the board of directors. Wang and Kelan (2013) show that the chance a female will be CEO increases when the amount of female directors in the company increases. The authors explain that this effect is caused by the critical mass as a result of having at least three female directors. This critical mass increases the chance that a female will be CEO. Wang and Kelan (2013) assume that the female quota has both a positive effect on increasing gender diversity in the boardroom and the chance that a female will be CEO. Besides, the authors argue that the increase of female board members in the board is a reaction of companies to legitimate themselves for the increasing demand from outside.
2.2.2 Token theory
Kanter (1977) argues that female directors in the board of directors are tokens because female directors are a minority in the top of the company (16.9 percent of female directors in the board of directors in the U.S. in 2014 according to Fortune (2014)). The token theory gives an explanation for the representation of minorities in the community. A minority of less than 15 percent is characterized by individuals that are not representing themselves but the category they belong to (Singh and Vinnicombe, 2004). A minority between the 15 and 30 percent is characterized as skewed meaning that the individuals are less isolated and give more social support to others (Singh and Vinnicombe, 2004). A minority between the 30 and 40 percent is tilted meaning that the population is still a minority but of such a size that it has substantial power (Singh and Vinnicombe, 2004).
It is for the minority (the token individuals) important that they behave in such a way that they fit within the top management (Singh and Vinnicombe, 2004). According to Kanter (1977) do female directors (token individuals) need to have a public face at work and hide their private face for succeeding to fit within the top management. Kanter (1977) argues that these two faces are needed because female directors get a lot of pressure from male directors to perform as best as possible. The low percentage of female directors in the board makes them more visible in comparison to male directors what causes the pressure for female directors to perform (Kanter, 1977). The skewed minority of female directors can cause that male directors react in boundary heightening behavior. This means that male directors
10 heightens the boundaries for female directors what makes it difficult for female directors to exceed this boundary (Oakley, 2000). This boundary behavior is used by male directors because they are afraid that they lose some of their salary because female directors get less paid and might perform the same as male directors showing that male directors are overpaid (Oakley, 2000).
2.3 Explanations for board gender diversity
2.3.1 Person-centered explanations
Person-centered explanations are the individual factors that explain why women and men differ in the execution of the job and in career choices (Helfat et al., 2006). Adams and Funk (2012) show in their study that there is a personal difference between men and women relating to the core values and the risk attitudes. Niessen and Ruenzi (2007) describe U.S. female managers as more risk averse, doing less extreme and more sensible investments than male managers. These authors also argue that female managers do not differ in performance in comparison with male managers but get less paid.
Contradicting to the findings of the study of Niessen and Ruenzi (2007) is the study Adams and Funk (2012). Adams and Funk (2012) argue that Swedish female directors are not risk averse but are more risk loving than male directors. Besides, Adams and Funk (2012) argue that Swedish female directors are less power oriented and more universally oriented than male directors. This shows that a difference in the behavior of female directors exists and differs due to nationality. A good example of the difference in characteristics of female
directors caused by nationality is illustrated in the study of Singh and Vinnicombe (2004). In an interview with a successful U.S. female director was asked why U.S. female directors are so successful in the U.K. The female director answered that in the U.S. an environment of high expectations exists within the society that keeps you to be focused and to perform as best as possible to reach the top (Singh and Vinnicombe, 2004).
Another difference between female and male directors is that female directors might lack the necessary experience to fulfill the function (Oakley, 2000). Oakley (2000) argues that a lot of women do not have the needed line experience to make it to a senior management position. Besides, women and men differ in communication styles. The communication style of female managers is characterized as more friendly and less assertive than the
11 2.3.2 Situation-centered explanations
The situation-centered explanations are the factors related to the organization and the group that can cause the differences in appointments and promotions between men and women (Helfat, et al., 2006). Farrel and Hersch (2005) argue that an increase of female directors in the board of directors is negatively related to the female directors already in the board. It is more likely that when a female board member leaves, this board member will be replaced by another female (Farrell and Hersch, 2005). A reason for this action of the company is that it wants to legitimate itself against the society for otherwise the decreasing amount of female board members.
Farrell and Hersch (2005) argue that there is a relationship between firm performance and the appointment of female board members. If the firm performance is good, companies are more likely to appoint female board members. Besides, Oakley (2000) argues that the corporate practices such as training, career development and compensation have an important influence on gender diversity in boards. These corporate practices are important for the prevention of women to break through the glass ceiling if companies do not offer these corporate practices to women (Oakley, 2000).
The size of the company has also an important relationship with the appointment of female board members. If the size of the company increases than it is more likely that female board members will be appointed (Farrell and Hersch, 2005, Harrigan, 1981). Adams and Ferreira (2009) argue that bigger companies are more likely to appoint female board members because the society sees these companies as role models for smaller companies and the bigger companies want to legitimate themselves or the fact that female board members are more attracted by bigger size companies.
2.3.3 Social system-centered explanations
Social system-centered explanations are factors relating to the society that can cause the differences in appointments and promotions between men and women (Helfat et al., 2006). According to Brammer, Millington and Pavelin (2007) is board gender diversity influenced by the external environment of the company. Besides, the authors argue that the customers of the company play a more important role in board diversity than the workforce of the industry (Brammer et al., 2007). A reason for this can be that companies want to show to their
12 customers that they are just as gender diverse as the customers (Brammer et al., 2007).
An important cultural issue related to board gender diversity is stereotyping (Oakley, 2000; Singh and Vinnicombe, 2004). According to Singh and Vinnicombe (2004) is
stereotyping an important barrier for women to reach the top of the company. In the studies of Broverman, Vogel, Broverman, Clarkson and Rosencrantz (1972) and Heilman, Block, Martell and Simon (1989) male managers were asked for their opinion about female
managers. Both studies reveal that female managers are characterized as: “less self-confident, less analytical, less emotionally stable, less consistent and possessing poorer leadership abilities than male managers” (Oakley, 2000, pp. 326). Besides, the looks, clothing and voice of women are stereotyped by men (Oakley, 2000). To beat this stereotyping, women have to camouflage these female characteristics to increase their credibility towards men (Jamieson, 1995).
The stereotyping of women by men has such an influence on women that they behave different in their decision making than when they are not stereotyped (Carr and Steel, 2010). Carr and Steel (2010) argue that women who are negatively stereotyped are more loss and risk averse than men and women who are not being stereotyped. This behavior can be explained by the ego depletion causing stereotyped women to behave more on intuition than on rationality (Carr and Steel, 2010).
2.4 Hypothesis development
The board of directors is responsible for the monitoring of the management and gives the management strategic advice (Baysinger and Butler, 1985). One could argue that the influence of board gender diversity could be related to the company’s strategy and the
execution of the company’s operations. Companies with complex operations benefit from the advice of the board of directors (Anderson, Reeb, Upadhyay and Zhao, 2011). These authors argue that complex operations could be an unrelated company that operates in different industries what makes it difficult for managers to understand all these industries. Complex companies can also be companies that operate in complex environments where the managers face troubles to know and meet the expectations of customers. It is therefore not surprising that a manager who faces operational complexity requests for advice from the board of directors. The quality of advice from the board of directors increases when the board is heterogeneous (Anderson, et al., 2011). A heterogeneous board can be characterized by members from different backgrounds, education, gender, ethnicity, etc. This means that
13 gender diversity itself influences the advice from the board of directors because of the
difference in problem thinking between men and women (Anderson et al., 2011).
However, a heterogeneous board could also lead to a conflict of interest (Van der Walt, Ingley, Shergill and Townsend, 2006; Goodstein, Gautam and Boeker, 1994). In their study, Van der Walt et al. (2006) examine the relationship between strategic complexity and board gender diversity for public-listed companies in New Zealand. The study reveals that companies with high strategic complexity in combination with high board gender diversity are performing worse than companies with low strategic complexity in combination with high board gender diversity (Van der Walt et al., 2006).
Nevertheless these contradicting results, one could propose that board gender diversity has a negative relationship with operational complexity. Complex operations need a lot of information sharing between board members (Faleye, 2007) but because of the stereotyping of women by men, one could propose that men are not willing to share their information with women. Men could propose that women are not capable (i.e. women are risk averse or do not have the skills to deal with complex situations) enough to deal with the information what in the end can lead to a decrease in the performance of the board of directors. To test whether this negative relationship between board gender diversity and complex operations exists, the hypothesis will be:
H1:There is a negative relationship between the company’s complexity of operations and board gender diversity.
The risk taking behavior of a company is in some cases crucial for the survival of the company or to gain competitive advantages. The company’s risk taking behavior is also present in the board of directors. The study of Niessen and Ruenzi (2007) shows that U.S. female board members are less willing to bear risk than male board members. This is supported by the study of Baixauli-Soler, Belda-Ruiz and Sanchez-Marin (2015). In this study, the authors show that female board members behave more conservative related to risk taking. This conservative behavior of women is less favorable for companies that need risks taking behavior to survive or to differentiate from competitors (Baixauli-Soler et al., 2015). For these companies, women are less likely to be appointed because this will reduce the company’s risk appetite.
Based on these results, I would like to test if these companies who need more risk taking behavior would be disadvantaged if they have female board members on their boards.
14 This could mean that it would be better for these companies to not follow the uniformity of corporate governance structures in the case of board gender diversity. The hypothesis to test the assumption will be:
H2: There is a negative relationship between the riskiness of the company and board gender
diversity.
Because of the low percentage of women in the board of directors women could see themselves as tokens and would like to represent the “woman” in the board of directors. However, because of this small amount of female directors, women might face that they are being stereotyped by men. This stereotyping can result in less risk and loss taking behavior of these women because they will behave more on intuition than on rationality (Carr and Steele 2010). I expect that this intuition based behavior of women have a negative influence on loss-making companies. I propose that these women are less able to handle losses because of their conservative behavior regarding losses and risks. Besides, I expect that in the context of stereotyping, women are less able to handle losses because they are less emotional stable than men. For a loss-making company is risk taking behavior (i.e. reorganization) sometimes needed to handle the losses but women are less willing to engage in such actions (i.e. afraid that this action causes more losses) or are less able to do this because of emotional motives (i.e. not willing to dismissal employees). The hypothesis to test this assumption will be:
H3: There is a negative relationship between loss-making companies and board gender diversity.
As mentioned before, a lot of academic papers are dedicated to the relationship between board gender diversity and firm performance. The study of Bilimoria (2006) show that board gender diversity has a positive relationship with financial performance for U.S. public-listed companies whereas the study of Carter et al. (2010) finds no evidence for this relationship.
However, I am interested what the financial consequences are for the companies that deviate in my empirical model I use for predicting board gender diversity. This means that I would like to know if a difference exists in the financial performance between companies that do not deviate from this empirical model and the companies that do deviate. When there is no considerable difference in performance, than it does not matter whether or not a company
15 deviates from this empirical model. I mean with considerable that an average skilled person could argue that the difference in performance is not almost the same). To find out if this relationship exists, I propose that the companies that deviate from my empirical model perform less than these that do not deviate since I think that my empirical model is predictive enough to explain the relationship between board gender diversity and financial performance. The hypothesis to test this assumption will be:
H4: The companies that deviate in board gender diversity according to my empirical model that I use for predicting board gender diversity perform considerable less than the companies that do not deviate.
16 3 Methodology
3.1 Measurement of board gender diversity
Board gender diversity is measured by the variable WOMEN. This variable is determined by dividing the number of female directors in the board by the total number of board members. This variable shows how much percent of female directors are represented in the board.
3.2 Measures for operational complexity
The first measure for operational complexity is ASS/SALES. I use this measure for measuring assets intensity (proportion assets divided by net sales). This measure shows how much assets a company needs to generate its sales. The higher the assets intensity is, the higher is the complexity of the company.
The second measure for operational complexity is INDUSTRY. These are the different industries a company operates in and can be used as a measure of operational complexity (Bradshaw, Miller and Serafeim, 2009). It is argued that it is more likely that complex companies operate in more than one industry. This measure is determined by using industry dummies for the different industries in my sample.
The third measure for operational complexity is PPE/ASS. This is the proportion of dividing property, plant and equipment (PPE) by total assets (Faleye, 2007; Anderson et al., 2011). According to Faleye (2007) have companies with more complex operations less PPE in comparison with the total assets and rely more on intangible resources to operate such as high qualified employees.
3.3 Measures for risk taking behavior
The risk taking behavior in this study will be related to a company’s financial risks. Chava and Purnanandam (2010) argue that risky companies are characterized by a high level of leverage, high amounts of short term debt and low amounts of cash holdings. I use the following four measures described below to test the risk taking behavior of a company. The first measure for risk taking behavior is LEV. This is known as leverage and is determined by the proportion total liabilities divided by total assets (Ahern and Dittmar, 2012).
17 The second measure is D/E which is the proportion debt to equity. I use this measure since risky companies have more debt than equity resulting in a higher debt to equity ratio. The third measure is CASH which are the cash holdings of the companies. The ratio for cash holdings is determined by the short term investments and cash divided by total assets (Chava and Purnanandam, 2010).
The fourth measure is DMAT which is debt maturity. Debt maturity is determined by dividing long term debt by total debt (Chava and Purnanandam, 2010). Debt maturity shows whether a company has relative more short term or long term debts. Long term debt in this study is defined as all the debt with a maturity longer than one year.
3.4 Measure for loss-making companies
The measure LOSS is the only measure I use for determining whether a company is loss-making or not. The measure LOSS is a dummy variable which is 0 if the company has a positive EBITDA and is 1 if the company has a negative EBITDA and is therefore loss-making. I use only one measure since this measure is based on EBITDA and a lot of performance measures use EBITDA as their basis such as return on investment.
3.5 Measures for subsequent firm performance
The first measure for subsequent firm performance is ROI t+1. This measure measures the return on investment (ROI) (net income divided by invested capital) of the following year based on actions taken in the prior year.
The second measure is ROS t+1. This measure measures the return on sales (ROS) (net income divided by net sales) of the following year based on actions taken in the prior year.
I use these measures because Miller and Triana (2009) argue that the combination of ROI and ROS is a good overall performance measure because ROI is comprehensive and ROS is not biased by accounting methods. Besides, the advantage of using two measures in the regressions is that it gives more reliable results since the usage of one measure can be more affected by potential bias.
18 3.6 Control variables
For this study I use five control variables that controls for the potential effects between board gender diversity and the chosen independent variables described above.
The first control variable is FSIZE. This variable is the firm size that is determined by the natural logarithm of total assets. The firm size is used as control variable because larger companies can behave slower to changes (i.e. operational complexity) (Greve, 2011). Besides, as mentioned before, a positive relationship exists between board gender diversity and the size of the firm (Farrell and Hersch, 2005).
The second control variable is BSIZE. This variable is the board size and is determined by the total number of directors in the board of a company. Board size is used as control variable because Anderson et al, (2011) argue that it is more likely that larger boards are more gender diverse than smaller boards.
The third control variable is GROWTH. This variable is related to the growth opportunities that are determined by the annual growth rate of net sales. The growth rate is used as control variable because according to Lehn, Patro and Zhao (2009) have growth opportunities a significant effect on the composition of the board of directors. Companies with high growth opportunities are often operating in more volatile business environments and need therefore a corporate governance structure that can adapt to these business environments (Lehn et al., 2009).
The fourth control variable is IND-DIR which is related to the proportion of independent board members in the board of directors. The variable is determined by the proportion of independent directors to the total board members. The proportion of
independent board members is used as control variable since boards consisting of independent directors are more likely to be heterogeneous (Anderson et al., 2011).
The fifth control variable is ROA which is the return on assets. Anderson et al. (2011) argue that ROA (EBITDA/total assets) controls for the effect that past performance can have on the relationship between the current performance and the composition of the board.
3.7 Empirical models
After I define my dependent, independent and control variables, I develop my empirical models. The first empirical model is used to predict why female directors are appointed in the board of directors. The second empirical model is used to explain what the influence is of the
19 mismatch in the first empirical model on subsequent financial performance.
This first empirical model is as follows:
WOMEN = ɑ0 + ɑ1ASS/SALES + ɑ2INDUSTRY + ɑ3PPE/ASS + ɑ4LEV + ɑ5D/E +ɑ6CASH
+ ɑ7DMAT + ɑ8LOSS + ɑ9FSIZE + ɑ10BSIZE + ɑ11GROWTH+ ɑ12IND-DIR +
ɑ13ROA+ ε
WOMEN = Proportion of female directors to the total number of directors in the board ASS/SALES = Proportion total assets to net sales
INDUSTRY = Dummy variable with different numbers for the different industries a company operates in
PPE/ASS = Proportion property, plant and equipment (PPE) to total assets LEV = Proportion total liabilities to total assets
D/E = Proportion debt to equity
CASH = Proportion short term investments and cash to total assets DMAT = Proportion long term debt to total debt
LOSS = Dummy variable that is 0 if EBITDA is positive and 1 if EBITDA is negative FSIZE = Natural logarithm of total assets
BSIZE = Total number of directors in the board GROWTH = Annual growth rate of net sales
IND-DIR = Proportion independent directors to total directors in the board ROA = Proportion EBITDA to total assets
For the operational complexity hypothesis (H1) I propose that ɑ1 has a negative
coefficient since I propose that companies with less assets intensity are more likely to be board gender diverse. Furthermore, I propose that ɑ2 has a negative coefficient because if a
company operates in more than one industry it is more complex and less likely to appoint female directors in their boards. I propose that ɑ3 has a positive coefficient since I propose
20 more likely to appoint female directors in the board.
For the risk taking behavior hypothesis (H2) I propose that when female directors are appointed in the board the riskiness of the company decreases. I propose that ɑ4 and ɑ5 have
negative coefficients because the debt level decreases when female directors are appointed because of their conservative behavior. Furthermore, I propose that ɑ6 has a positive
coefficientand ɑ7 has negative coefficient since riskier companies have less cash holdings and
more short term debt. However, I propose that if female directors are appointed the riskiness of the company decreases.
For the loss-making companies hypothesis (H3) I propose that ɑ8 has a negative
coefficient since the appointment of female directors will have a negative influence on the financial performance of the firm.
The second empirical model is as follows:
ROI t+1 or ROS t+1 = ɑ0 + ɑ1 WOMEN or MISMATCH+ɑ2FSIZE + ɑ3BSIZE +
ɑ4GROWTH + ɑ5IND-DIR + ɑ6ROA + ε
ROI t+1 = net income next year divided by invested capital next year ROS t+1 = net income next year divided by net sales next year
WOMEN = Proportion of female directors to the total number of directors in the board MISMATCH = Residuals in the variable WOMEN in my empirical model I use for
predicting board gender diversity FSIZE = Natural logarithm of total assets BSIZE = Total number of directors in the board GROWTH = Annual growth rate of net sales
IND-DIR = Proportion independent directors to total directors in the board ROA = Proportion EBITDA to total assets
For the mismatch hypothesis (H4) I do two regressions. In the first regression, I use the variable WOMEN as independent variable whereas I use the variable MISMATCH in the second regression. I propose that ɑ1 of the variableWOMEN is considerable higher than the
21 4 Data and model
4.1 Sample
The sample consists of U.S. public-listed companies on the S&P 1500 for the years 2006 till 2010. I use this time period because in this time period serious steps are taken by the
governments worldwide (i.e. female quotas) to increase the appointment of women in the board of directors. Besides, I would like to know if the appointment of female directors is changed due to the financial crisis.
I begin my data collection with collecting governance data from the database “ISS”. I collect data regarding the gender and the status of independence of the board members. Besides, I collect data regarding the amount of seats in the board per company. The financial data is collected from the database “Compustat”. I collect financial data regarding the
financial structure of the company such as debt, equity and invested capital but also net income, assets and net sales. After merging and removing incomplete data, the sample
consists of 947 unique firms yielding 5,375 firm years. Having this sample as starting point, I remove the financial services industries (SIC 6000-6999) because companies in these
industries are more likely to have a divergent financial structure than the companies in the other industries. I also remove the companies that are not represented in all the years in the chosen time period. The final step I take is winsorizing the top and bottom 1% of the values of the variables ASS/SALES, D/E, ROA and GROWTH because otherwise could extreme observations in these variables bias my results. This gives me in the end a sample of 365 unique firms yielding 2,190 firm years.
4.2 Descriptive statistics
The distribution of the observations per industry in the sample is shown in table 1. Table 1 shows that the numbers of observations per industry are not equally distributed. The sample is dominated by the manufacturing industry that covers 54 percent of all observations in this sample followed by the transportation and utilities industry that covers 21 percent of all observations. The services industry and wholesale & retail trade industries follows with 10 and 9 percent of all observations respectively. The observations of the agriculture, forestry and fishing industry should be interpreted with caution as a result of the small amount of observations. The total sample consists of 14,320 observations.
22 TABLE 1
Distribution of observations per industry
SIC Industry Number of
observations
Percentage of the sample
0-999 Agriculture, Forestry & Fishing 75 1% 1000-1799 Mining & Construction 792 5%
2000-3999 Manufacturing 7,674 54%
4000-4999 Transportation &Utilities 3,039 21% 5000-5999 Wholesale & Retail Trade 1,232 9%
7000-8999 Services 1,508 10%
Total number of observations 14,320
The percentage of women in the board per fiscal year is shown in table 2. Table 2 shows that the percentage of women in the board was increasing in the period prior to the financial crisis. In this period the percentage of women in the board went up from 12.8 percent in 2005 up to 14.1 percent in 2007. After the first year of the financial crisis (2008), the percentage of female directors in the board decreased. In 2008 was the average amount of female directors the highest with 14.6 percent and from then on was this percentage
decreasing in the following years to 13.9 percent in 2010. This means that during the financial crisis companies did not continue to increase the percentage of women in the board but
returned to their old behavior by favoring men above women.
TABLE 2
The average percentage of women in the board per fiscal year
Fiscal year Percent of women in the board
2005 12.8% 2006 13.3% 2007 14.1% 2008 14.6% 2009 14.4% 2010 13.9%
23 In table 3 are the percentages of women in the board per industry per year shown. For example the manufacturing industry has an increase of women in the board from 2005 till 2007 and was stable in 2008 but from then on the percentage of women in the board declined. The table shows that on average the percentage of women in the board per industry is not considerable different from each other and therefore a distinction between industries for the reminder of the study will not be made.
TABLE 3
Percentage of women in the board per industry per year
SIC Industry 2005 2006 2007 2008 2009 2010 Aver.
0-999 Agriculture, Forestry & Fishing
16.7% 16.7% 8.3% 8.3% 15.4% 14.3% 13.3% 1000-1999 Mining & Construction 10.0% 11.7% 11.9% 13.0% 12.8% 12.8% 12.0% 2000-3999 Manufacturing 12.2% 12.9% 14.3% 14.3% 14.1% 13.4% 13.5% 4000-4999 Transportation &Utilities 14.7% 14.7% 14.8% 16.6% 16.1% 15.3% 15.4% 5000-5999 Wholesale & Retail Trade 11.7% 10.9% 12.0% 12.3% 11.4% 12.8% 11.9% 7000-8999 Services 14.2% 14.9% 14.5% 15.1% 15.3% 15.4% 14.9%
The descriptive statistics of the variables used for this study are shown in table 4. The average of the variable WOMEN is 13.8 percent meaning that the average amount of female directors in the board is 13.8 percent. The average company has an assets/net sales
(ASS/SALES) proportion of 1.39 and a PPE/assets (PPE/ASS) proportion of 0.33, leverage of 0.52 and a debt/equity (D/E) proportion of 0.83.
Table 4 shows that the financial characteristics of the companies are very divergent. For example, the proportion debt/equity (D/E) is on average 0.66 while the median is 0.46 and the standard deviation is 0.81. This difference in financial characteristics can be caused by the different industries in this sample. For example, a company in a service industry needs a lot less assets to operate than a manufacturing company. In contrast to the financial
characteristics is the difference in corporate governance characteristics between companies less present. The average board size (BSIZE) is 7.2 while the median is 7.0 and the average
24 board has 5.2 independent directors (IND-DIR) while the median has 5.0 independent
directors.
TABLE 4
Descriptive statistics
Variable Mean Std. Dev. Q1 Median Q3 No.
observations WOMEN 13.8% 12.4% 0 14.3% 22.2% 14,320 ASS/SALES 1.39 0.84 0.82 1.18 1.72 14,320 INDUSTRY 3.63 1.09 3.00 3.00 4.00 14,320 PPE/ASS 0.33 0.24 0.12 0.26 0.53 14,320 LEV 0.52 0.19 0.39 0.53 0.68 14,320 D/E 0.66 0.81 0.14 0.46 0.88 14,320 CASH 0.12 0.13 0.02 0.07 0.17 14,320 DMAT 0.90 0.17 0.88 0.96 1 12,545 LOSS 0.09 0.28 0 0 0 14,320 FSIZE 8.1 1.5 7.0 8 9.1 14,320 BSIZE 7.2 2.1 6.0 7.0 8.0 14,320 GROWTH 5.8% 16.9% -2.9% 6.0% 14.0% 11,765 IND-DIR 5.2 1.9 4.0 5.0 6.0 14,320 ROA 14.5% 6.9% 9.6% 13.7% 18.8% 14,320
INDUSTRY is a dummy variable that has values between 1 till 6 based on the industries shown in table 3
Like the study of Faleye (2007), I divide the sample in two groups to make the
comparison between not gender diverse and gender diverse boards possible. The first group is related to the boards that are not gender diverse. These boards consist of only men. The second group is the gender diverse boards that consist of a mix of gender. In table 5 is the comparison between these two groups shown.
Companies with gender diverse boards have lower cash holding (CASH) that might be caused due to higher levels of debt (D/E) and leverage (LEV) than the companies without gender diverse boards. According to Chava and Purnanandam (2010) are these characteristics of the board gender diverse companies the characteristics of riskier companies. Besides, board gender diverse companies are less loss-making (LOSS) than not board gender diverse
25 and ROA) than companies without board gender diversity. Regarding the difference in
governance characteristics between the two groups are board gender diverse companies more likely to be larger firms (FSIZE) with larger boards (BSIZE) consisting of more independent directors (IND-DIR). These characteristics of lower performance from larger companies with larger independent boards of board gender diverse companies are supported by Adams and Ferreira (2009) who reveals these findings in their study.
TABLE 5
Comparison of firm characteristics between boards without gender diversity versus boards with gender diversity
Variable Not gender diverse
board
Gender diverse board T-Test P-value
Mean Mean ASS/SALES INDUSTRY PPE/ASS 1.38 3.63 0.33 1.43 3.67 0.35 -2.05 -1.39 -2.70 0.04** 0.17 0.007*** LEV 0.52 0.57 -12.69 0.000*** D/E 0.64 0.80 -7.96 0.000*** CASH 0.12 0.10 6.88 0.000*** DMAT 0.90 0.90 -2.37 0.018** LOSS 0.09 0.08 0.67 0.50 FSIZE 8.0 8.5 -12.46 0.000*** BSIZE 7.1 7.5 -7.89 0.000*** GROWTH 0.06 0.05 3.46 0.001*** IND-DIR 5.17 5.68 -11.69 0.000*** ROA 14.5 14.3 1.15 0.25 No. observations 12,339 1,981
The p-values with significant values are indicated by ***, ** or * indicate significance at 1%, 5% or 10% levels (two-tailed) respectively.
The comparisons of firm characteristics between board gender diverse and not board gender diverse companies are also statistically tested by using the paired sample t-test. The results of the t-tests show that almost all the variables are significant different between the two groups. Therefore, these significant results in the variables due to gender give a good
26 basis for testing the hypotheses. However, table 5 also shows that there is no significant difference between board gender diverse and not board gender diverse companies when it comes to industries (INDUSTRY), loss-making (LOSS) and ROA. Based on the results of the variable LOSS is the loss-making company hypothesis (H3) already rejected because there is no significant difference in loss-making between board gender diverse and not board gender diverse companies.
4.3 The correlation matrix
Faleye (2007) mentions in his paper that a main concern is the correlation among the
independent variables what makes them interdependent and could therefore bias the results of the regressions. To figure out if this concern of Faleye (2007) is applicable to the independent variables I use for testing the hypotheses, I make a correlation matrix. This correlation matrix is shown in table 6 and consists of the independent variables and the dependent variable I use in this study. By using these variables, I examine whether the independent variables are correlated with the dependent variable and if the independent variables are correlated to each other. If this latter is the case than my results of the regressions could be biased.
The first number in the table shows the Pearson’s correlation coefficient that is a measure used for identifying correlation. The coefficient has a value between +1 and -1. The closer the coefficient is to +1 or -1 the more correlated is the variable with another variable. At the first sight are the variables not strong correlated with each other since most variables have values of less than -0.59 and +0.59 meaning that the variables are at most moderate correlated (Evans, 1996). However, the variables D/E and LEV are strong positive correlated with each other (+0.69). By using these two variables together, the variables could bias the results of the regressions.
The second number (between the hedges) shows the p-value of the correlation. The figures in the table shows that almost all the correlations are significant. Therefore, by using these variables the results of the regressions should be interpreted with caution.
27
TABLE 6
The correlation matrix with the independent variables and the dependent variable
WOMEN ASS/SALES INDUSTRY PPE/ASS LEV D/E CASH DMAT LOSS
WOMEN 1.0000 ASS/SALES 0.0470*** (0.0000) 1.0000 INDUSTRY 0.0322*** (0.0001) -0.0638*** (0.0000) 1.0000 PPE/ASS 0.0622*** (0.0000) 0.3822*** (0.0000) -0.0808*** (0.0000) 1.0000 LEV 0.2867*** (0.0000) 0.1770*** (0.0000) 0.1051*** (0.0000) 0.3069*** (0.0000) 1.0000 D/E 0.1995*** (0.0000) 0.2596*** (0.0000) 0.0681*** (0.0000) 0.2580*** (0.0000) 0.6904*** (0.0000) 1.0000 CASH -0.1433*** (0.0000) -0.0551*** (0.0000) -0.0151*** (0.0000) -0.4249*** (0.0000) -0.4728*** (0.0000) -0.2676*** (0.0000) 1.0000 DMAT 0.0563*** (0.0000) 0.0855*** (0.0000) -0.0639*** (0.0000) 0.1593*** (0.0000) 0.2043*** (0.0000) 0.0973*** (0.0000) -0.2265*** (0.0000) 1.0000 LOSS -0.0154*** (0.0661) 0.1067*** (0.0000) -0.0817*** (0.0000) -0.0123 (0.1412) 0.0218*** (0.0091) 0.0862*** (0.0000) 0.0947*** (0.0000) -0.0338*** (0.0002) 1.0000
28
The p-values with significant values are indicated by ***, ** or * indicate significance at 1%, 5% or 10% levels (two-tailed) respectively.
4.4 The relationship between board gender diversity and firm characteristics
After these comparisons of variables between board gender diverse and not board gender diverse companies and testing for correlations by using a correlation matrix, I start with the OLS regressions. These OLS regressions test whether a relationship exists between board gender diversity and firm characteristics. I use two OLS regressions to test this. The first regression (column I in table 7) test the relationship between board gender diversity and the independent variables described above. In the second regression (column II in table 7) I add the control variables in the regression to examine what the effect of the control variables are on the results of the regression. By using these two regressions I examine whether the results are arbitrary (column I) or can be explained by adding the control variables in the regression (column II).
The results of the regression (table 7) show that the first regression (column I) has a much lower R-squared (0.0721) than the second regression (0.1250). This shows that by adding the control variables in the empirical model, the data fits better in the model. The Variance Inflation Factors (VIFs) in both regressions are quite low showing that there is no multicollinearity problem. In the first regression has the variable LEV the highest VIF with a value of 2.14 and in the second regression has the variable IND-DIR the highest VIF with a value of 3.86. It is argued that a VIF above the value of 5 is worrisome.
When comparing the two regressions, the variables PPE/ASS and LEV have in both regressions coefficients with significant p-values. The variable PPE/ASS tells us that complex companies are more likely to appoint female directors due to the decrease in the PPE to assets proportion. Therefore, this variable gives no support for the operational complexity
hypothesis (H1). The variable LEV shows that companies with more leverage are more likely to appoint female directors in the board. These companies with more leverage are therefore more likely to be risky. Therefore, the variable LEV gives no support for the risk taking behavior hypothesis (H2).
The variable ASS/SALES has a negative coefficient with a significant p-value in the second regression showing that the results are not random since the relationship can be explained in combination with control variables. If a company has more assets intensity
29
TABLE 7
The results of the OLS regressions using board gender diversity as dependent variable
Variable I II ASS/SALES -0.0010 (0.457) -0.0119*** (0.000) INDUSTRY -0.0017 (0.139) -0.0000 (0.971) PPE/ASS -0.0113** (0.032) -0.0122** (0.033) LEV 0.1996*** (0.000) 0.1286*** (0.000) D/E -0.0009 (0.649) 0.0032 (0.112) CASH -0.0137 (0.216) -0.0351*** (0.003) DMAT 0.0001 (0.987) -0.0201** (0.017) LOSS -0.0085** (0.042) 0.0071 (0.124) FSIZE 0.0152*** (0.000) BSIZE -0.0064*** (0.000) GROWTH -0.0409*** (0.000) IND-DIR 0.0141*** (0.000) ROA 0.0062 (0.767) R-squared 0.0721 0.1250 Highest VIF 2.14 (LEV) 3.86 (IND-DIR) No. observations 12,545 10,294
30
WOMEN = proportion of female directors to the total number of directors in the board, ASS/SALES = proportion total assets to net sales, INDUSTRY = dummy variable with different numbers for the different industries a company operates in, PPE/ASS = proportion property, plant and equipment (PPE) to total assets, D/E = proportion debt to equity, LEV = proportion total liabilities to total assets, CASH = proportion short term investments and cash to total assets, DMAT= proportion long term debt to total debt, LOSS = dummy variable that is 0 if EBITDA is positive and 1 if EBITDA is negative, FSIZE = natural logarithm of total assets, BSIZE = total number of directors in the board, GROWTH = annual growth rate of net sales, IND-DIR = proportion independent directors to total directors in the board, ROA = proportion EBITDA to total assets. The p-values with significant values are indicated by ***, ** or * indicate significance at 1%, 5% or 10% levels (two-tailed) respectively. The p-values in columns I and II are adjusted for heteroskedasticity by using robust standard errors.
(ASS/SALES) than the likelihood of appointing female directors decreases. Based on this result it is proposed that board gender diverse companies have less assets intensity
(ASS/SALES) what makes them less complex and gives support for the operational
complexity hypothesis (H1). The number of industries a company operates in (INDUSTRY) has no relationship with board gender diversity in both regressions. This is also the case for the variable D/E which has also no significant coefficients. The variable CASH has a negative coefficient in both regressions from which they are significant in the second regression. The results of the variable CASH are like the results of the variable ASS/SALES not random. So, when a company is board gender diverse than the cash holdings decreases what makes the company riskier and gives no support for the risk taking behavior hypothesis (H2). The variable DMAT has also a significant p-value in the second regression with a negative coefficient. This means that when a company is board gender diverse it has more short term debts and is therefore riskier than a not board gender diverse company. This gives also no support for the risk taking behavior hypothesis (H2). The variable LOSS is only significant in the first regression that shows that this significance is caused due to randomness.
At the first sight do these results look strange since women are known for their risk avoiding behavior but why are female directors appointed in riskier companies? This behavior can be explained by the glass cliff theory. The glass cliff refers to the phenomenon that
companies prefer female directors over male directors in leadership positions under circumstances with more pressure due to risk or failure (Francoeur, Labelle and Sinclair-Desgagné, 2008). An explanation for the glass cliff is that if female directors are appointed and the performance is still decreasing than the female directors are more willing to take their responsibility for the bad results and they are better able to manage people under stressful situations (Ryan, Haslam, Herby and Bongiorno, 2011).
31 diverse companies are more likely to be larger firms. The variable BSIZE has a negative coefficient with a significant p-value showing that board gender diverse companies have smaller boards. This result is contradicting to the findings of Anderson et al. (2011) who argue that board gender diverse boards are larger than not gender diverse boards. The variable GROWTH has a negative coefficient with a significant p-value revealing that board gender diverse companies have a lower growth rate of annual net sales than not board gender diverse companies. According to Lehn et al. (2009) do these companies with lower growth rates have
to adapt less quickly to their business environment since this environment is less volatile. Moreover, board gender diverse companies have more independent directors
(IND-DIR) in their boards than not board gender diverse companies. This result is in line with the study of Anderson et al. (2011) that reveal that board gender diverse companies are more likely to consist of independent directors. The variable ROA has a positive coefficient but not a significant p-value and gives therefore weak evidence. This means that based on this
variable nothing can be said whether better performing companies (based on the ROA) are more likely to appoint female directors.
In general, based on the results in table 7 it is hard to support the operational
complexity hypothesis (H1). Only the variable ASS/SALES shows evidence that supports the hypothesis since the coefficient in the regression (column II) moves in the same direction as in my empirical model. The variable PPE/ASS shows contradicting results in the regression than that I have proposed in my empirical model and it gives therefore no support for the hypothesis. The variable INDUSTRY gives weak evidence in the regressions and gives therefore no support for the hypothesis.
The risk taking behavior hypothesis (H2) is not supported since the regressions show that board gender diverse companies are riskier than not gender diverse companies. The regressions show that board gender diverse companies have positive LEV and D/E coefficients (only in the second regression) and negative CASH and DMAT coefficients. These characteristics are according to Chava and Purnunandam (2010) the characteristics of risky companies. Moreover, the loss-making company hypothesis (H3) is not supported since the results of the first regression are biased by randomness and are therefore not significant in the second regression.
4.5 How to tackle the potential correlation among the variables in the regressions?
32 variables are correlated with each other. But could this correlation bias the results of the regressions? To answer this question, I use two methods. In the first method I do the same action as Faleye (2007) does in his study. I repeat the OLS regressions from table 7 and test if the significance from the previous regressions still exists when I do the regressions with each independent variable individually while using the same control variables. If the coefficients independent variables individually show significant results in relation with board gender diversity (WOMEN) that are in line with the previous regressions shown in table 7 (column II) than there is no correlation problem. In the second method I use principal component factor analyses to examine if combining the independent variables used for the operational complexity (H1) and risk taking behavior (H2) hypotheses would lead to different results.
The first attempt to tackle correlation is repeating the regressions for every independent variable individually while keeping the dependent and control variables the same. The individual regressions of the independent variables used for the operational complexity hypothesis (H1) show that the variable ASS/SALES is significant (p-value is 0.0001). The variable PPE/ASS is not significant (p-value is 0.095) meaning that this variable is interdependent with the other independent variables because it is significant in the OLS regressions from table 7. Surprisingly, the variable INDUSTRY has a significant coefficient (p-value is 0.0001) in the individual regression meaning that the interdependences between the independent variables causes that this variable is not significant when doing the OLS regressions from table 7. The independent variables used for the risk taking behavior
hypothesis (H2) have the same interdependence problem. The variables LEV, D/E, CASH are all significant (p-value is 0.000) when doing the individual regressions but the variable
DMAT is not significant (p-value is 0.792) meaning that this variable is interdependent because it is significant in the OLS regression from table 7 (column II).
The second attempt is doing the same regression as in table 7 (column II) but adjust the independent variables used for the operational complexity (H1) and risk taking behavior (H2) hypotheses by doing principal component factor analyses. I create a new variable for the independent variables regarding the operational complexity hypothesis (ASS/SALES,
INDUSTRY and PPE/ASS) called OP-COM and a new variable for the independent variables regarding the risk taking behavior hypothesis (LEV, D/E, CASH and DMAT) called RISK. I use these variables OP-COM and RISK in the regressions for testing the hypotheses instead of the independent variables I used in the regressions from table 7. I do this because I would like to examine whether the “combined” variables OP-COM and RISK can give me more
33 contradicting results.
The results of the regression are shown in table 8. The r-squared of the regression (0.1001) is slightly lower than these of the previous regression in table 7 (0.1250). The highest VIF in the regression is from the variable IND-DIR with a value of 3.77 showing that there is no need to worry for serious multicollinearity. The variable OP-COM gives evidence that board gender diversity has a negative relationship with the complexity of a company’s operations. This means that the results of this variable support the operational complexity hypothesis (H1). The variable RISK gives evidence that the riskiness of the company increases when female directors are appointed and gives therefore no support for the risk taking behavior hypothesis (H2). The variable LOSS and the control variables have comparable results as in table 7 which is logical since only the independent variables regarding the first two hypotheses are changed.
TABLE 8
The results of the OLS regression using board gender diversity as dependent variable and the adjusted independent variables due to principal factor component analyses
Variable I OP-COM -0.0036*** (0.002) RISK 0.0156*** (0.000) LOSS 0.0009 (0.841) FSIZE 0.0149*** (0.000) BSIZE -0.0057*** (0.000) GROWTH -0.0472*** (0.000) IND-DIR 0.0147*** (0.000) ROA 0.0203 (0.290)
34
R-squared 0.1001
Highest VIF 3.77
(IND-DIR)
No. observations 10,294
WOMEN = proportion of female directors to the total number of directors in the board, OP-COM = variable created after doing a principal component factor analysis for the variables ASS/SALES, INDUSTRY and PPE/ASS, RISK = variable created after doing a principal component factor analysis for the variables LEV, D/E, CASH and DMAT, LOSS = dummy variable that is 0 if EBITDA is positive and 1 if EBITDA is negative, FSIZE = natural logarithm of total assets, BSIZE = total number of directors in the board, GROWTH = annual growth rate of net sales, IND-DIR = proportion independent directors to total directors in the board, ROA = proportion EBITDA to total assets. The p-values with significant values are indicated by ***, ** or * indicate significance at 1%, 5% or 10% levels (two-tailed) respectively. The p-values in columns I and II are adjusted for heteroskedasticity by using robust standard errors.
Based on the two additional tests for correlation it can be said that the results of the first attempt show that correlation exists between the independent variables I use for the operational complexity (H1) and risk taking behavior (H2) hypothesis. However, in the
second attempt is this correlation solved by creating new variables due to principal component factor analyses. By using those two new variables in the regression, I find clear evidence to support the operation complexity hypothesis (H1) and evidence to not support the risk taking behavior hypothesis (H2).
4.6 The relationship between the mismatch in board gender diversity and subsequent
financial performance
After doing the previous regressions, I test the relationship between board gender diversity and subsequent financial performance. After I do this, I test what the relationship is of the mismatch (residual) of my empirical model I use for predicting board gender diversity on subsequent financial performance.When the coefficients of board gender diversity are almost equal to the coefficients of the mismatch in board gender diversity than the mismatch
hypothesis (H4) is not supported.
The previous regressions in table 7 are focused on the variability in the dependent variable WOMEN that is explained by the independent variables. However, there is always some variability in the dependent variable left that cannot be explained by the independent variables in the regression. This is the residual (ε) of my empirical model I use for predicting board gender diversity. I save this residual from the previous regression in table 7 (column II)
