Does Gender Diversity on the
Board of Directors Influence
Performance?
Student Number: 10062777 Specialization: Finance and Organization Field: Organizational Economics Supervisor: Silvia Dominguez MartinezJordi Dahlberg
In this paper I will examine whether the share of female directors has a significant influence on firm performance. I will use a sample of US listed firms, being the Standard and Poor’s 500 firms. I will use different performance measures for which I will show different results. On top of showing whether a relationship between female directors and firm performance exists at all, I will also show whether this relationship is positive or negative. The results show that not always a relationship exists between the share of female directors and firm performance. Mixed results are found, and therefore it is not proven that firms are worse of per se, when adding females to the board of directors, thereby diversifying the board in terms of gender.1. Introduction
Historically the share of females on the board of directors has been low. In 2007 only 14.8% of all the directors on the fortune 500 companies' boards were women (Adams and Ferreira, 2009). For other countries the percentages of women were not much higher, for example in Australia (8.7%), Japan (0.4%), Canada (10.6%) and Europe (8.0%). Though the share of women on boards of directors is increasing, the numbers are still small (Adams and Ferreira, 2009). Therefore, more and more governments are trying to force companies by law to increase the share of women in top‐ management positions (Daunfeldt and Rudholm, 2012). Norway started this in 2006, implementing a law that forced companies to have at least 40% males and females on the board of directors by January 2008. More countries in Europe, such as Iceland, Spain and France, followed Norway and implemented similar laws. Italy and Belgium are in the process of implementing such laws (Daunfeldt and Rudholm, 2012). There are lots of arguments in favor of a more diversified board, such as better decision‐making (Judge and Miller, 1991), better monitoring (Fairfax, 2005) and a better understanding of markets that themselves are diverse in terms of gender (Robinson and Dechant, 1997). However, there are also some arguments against a more diverse board in terms of gender. Jehn et al (1998) argue that a more diverse board in terms of gender causes more conflicts and employee turnover, while Williams and O'Reilly (1998) argue that a more diverse board could lead to more emotional conflicts.
Empirically seen there has been a lot of research on this topic, and the results shown in prior literature are inconsistent. Some studies find positive relationships between the share of female directors and firm performance (Carter et al, 2003; Erhardt et al, 2003; Campbell and Minguez‐Vera, 2008), some find negative results (Adams and Ferreira, 2009; Ahren and Dittmar, 2012; Daunfeldt and Rudholm, 2012). Others find no relationship at all (Farrel and Hersch, 2005; Ayuso et al, 2007). Most of the literature investigated the relationship based upon time‐periods that started and ended a while ago (e.g. Adams and Ferreira (2009) use 1996‐2003). Since Adams and Ferreira (2009) there have been several other investigations on the topic, but they were all done using European data (Campbell and Minguez‐Vera, 2008; Ahren and Dittmar, 2012; Daunfeldt and Rudholm, 2012). However the time‐periods of these studies did not consist of years after 2009 either. In my investigation I will consider the time‐period, 2008‐2012, using US data (S&P 500) like Adams and Ferreira (2009). By doing so I hope to contribute relevant information and evidence to the investigation and discussion on this topic.
The research question that will be investigated in this paper is: Does gender diversity on the board of directors influence performance? To answer this question ordinary least squares (OLS) regression will be used. In the regressions I will perform, I use five different independent variables,
regressing them on three different performance measures (dependent variables). These performance measures are the growth in sales, the return on total assets and the return on equity.
This paper shows that for certain performance measures gender diversity on the board of directors influences performance, while for other performance measures it does not. Moreover I will show that a higher share of female directors worsens performance for some of the performance measures used. The contradictory results I provide, show that the relationship between the share of female directors and firm performance is not clear. As Campbell and Minguez‐Vera (2008) said, different time‐periods, countries and performance measures tend to give different results. Still the investigation into this relationship is very important, since governments are trying to force companies to increase the number of female directors, while this might not be the best solution when looking at firm performance. As said, theoretically there are enough arguments in favor of a more diverse board. The empirical evidence however, shows mixed results.
The next section will provide, from a theoretical and empirical background, information on previous literature. It will describe positive and negative effects of the diversification of boards of directors in terms of gender. Then in section three I will describe the data and methodology I will use to conduct my investigation. In section four I provide the results of the regressions performed, which I will discuss in section five. Section six concludes. 2. Literature Review
Diversity in the workplace has become an important point of discussion among many companies. More and more firms have been incorporating training programs for their employees to emphasize the importance of diversity among the workplace (Holladay et al. 2003). But what is diversity and why is it important? And are there differences between males and females? In this section I will try to answer these questions. I will provide, from a theoretical and empirical perspective, background information about my topic. I will start in subsection 2.1 by describing diversity and some arguments in favor and against diversity of the board of directors. Then in section 2.2 I will provide evidence from prior literature on the relation between the share of females on the board of directors and firm performance. Lastly, in section 2.3 I will form and state my hypothesis based upon the evidence from section 2.2. 2.1 Diversity Diversity is the term used to describe a variety or multiformity. The definition of board diversity, as given by Carter et al. (2003), is the percentage of women, Asians, Hispanics and African Americans on the board of directors (Carter et al, 2003). Although I will look at gender diversity only in this paper, I
find it relevant to give the actual meaning of the term board diversity, since it makes clear there is more than gender diversity only, and the results of my investigation therefore will not give a complete overview of diversity on the board of directors, and its influence on performance. 2.1.1 Why is Gender Diversity Important? Many papers proposed arguments in favor of a more (gender) diverse board. The consultancy report by Catalyst (1995) for example found that increasing the share of female directors, thereby diversifying the board in terms of gender, increased the diversity in opinions. It furthermore found that a more diverse board provided more female role models and female mentors than a comparable number of males would have provided (Catalyst, 1995), which according to Smith et al (2006) gives companies a better image. They argue that people appreciate it when companies try to do something for the community, and by providing role models, the companies show themselves as more than a profit‐maximizing entity (Smith et al, 2006). According to Pelled et al (1999) males and females tend to have different cognitive biases. Therefore, they argue, women have different norms, beliefs and perspectives which makes the ideas in the board room broader and more diverse as the shares of both males and females are approximately equal (Pelled et al, 1999). Furthermore Jehn et al (1998) find that women bring more creativity and innovation to a group. According to both Eisenhardt (1989) and Judge and Miller (1991) processes such as decision making could be improved by more gender diverse boards, since there are more alternatives considered and evaluated.
Fairfax (2005) argues that the monitoring functions of the board are enhanced when more females are on the board of directors, since women are less likely to agree on extreme decisions, while they are also more likely to engage in higher quality analysis before decisions are taken (Fairfax, 2005). Bradshaw et al (1992) conclude that women are less attached to power and therefore have a better ability of sharing power, thereby contributing more to governance than men. Women furthermore tend to bring new strategic directions to a firm (Selby, 2000).
With a more gender diverse board, a firm might have a better understanding of markets that are diverse in terms of gender themselves (Robinson and Dechant, 1997). Levi et al (2008) examined the role of women during mergers and acquisitions. They found that independent female directors played a great role in the bid premium and the lower target abnormal announcement (Levi et al, 2008). Lastly, more diverse boards, in terms of gender, prevent corruption and fraud, according to Ramirez (2003), since females are likely to ask ''tougher'' questions to management.
2.1.2 Why Would Gender Diversity be Bad?
As shown in the former section there are a lot of arguments in favor of gender diverse boards. However, many papers give arguments why a gender diverse board would not be good. There are first
of all a lot of studies investigating the relationship empirically, but I will discuss these in section 2.2. In this section I will provide arguments against a more diverse board in terms of gender.
As previously mentioned Jehn et al (1998) find that women bring in more creativity and innovation to a group. However, they simultaneously find evidence of diversity causing more conflicts and employee turnover (Jehn et al. 1998). Williams and O'Reilly (1998) argue that a more diverse board in terms of gender could lead to more emotional conflicts. Furthermore, both Adams and Ferreira (2007) and Miller et al (1998) find that more gender diverse boards show lack of communication between males and females, while Lau and Murnighan (1998) find that diverse boards are more time‐consuming and therefore less effective in terms of decision making (Lau and Murnighan, 1998).
Farrel and Hersch (2005) find that gender has impact on the selection of directors which, according to them, is consistent with the idea that adding female directors to the board is done following internal or external calls for diversity (Farrel and Hersch, 2005). Their results argue that not only are women added to the board of directors solely based upon their qualifications, but also because there are numerous external calls for adding women (Farrel and Hersch, 2005). This pressure, according to Farrel and Hersch (2005), makes companies make decisions about directors they would not have made otherwise. This in turn could lead to the appointment of less qualified female directors, instead of better qualified male directors.
2.2 Women and Performance
In this section I will provide evidence from prior papers that investigated the relationship between diverse boards in terms of gender, and firm performance. First, I will describe two papers that are very relevant for my investigation. The first article I will discuss is an article by Adams and Ferreira. It investigates whether women in the boardroom influence firm performance (Adams and Ferreira, 2009). The second article I will discuss is an article by Hoogendoorn, Oosterbeek and van Praag (2013), who did a field experiment to test the relationship between share of females in management positions and firm performance. In the last section I will discuss other papers investigating the relationship, linking them to the first two articles discussed. 2.2.1 Women in the Boardroom and Their Impact on Governance and Performance Adams and Ferreira (2009) started their investigation because they were wondering if women had a positive influence on firm performance. In their introduction they state that by 2007 only 14.8% of the corporate board seats in the fortune 500 were held by women. On top they say that compared to other well industrialized countries like Australia, Canada, Japan and Europe the United States had the highest percentage of women on board seats. Their well‐structured paper discussed the basic facts
about female representation, examined the relation between (gender) diversity and board inputs and analyzed the relation between (gender) diversity and firm performance.
2.2.1.1 Data and Methodology
The sample they used for their investigation consisted of data for the S&P 500, S&P MidCap and S&P SmallCap firms, and ranged from 1996‐2003. They used data on directors, such as gender; age; number of other directorships; retirement status and their attendance on board meetings (should be >75% of total board and committee meetings), as well as financial data like CEO compensation and CEO tenure.
To secure their methodology Adams and Ferreira (2009) describe some omitted variables that could affect the selection of female directors and governance choices, which could lead to correlations between gender diversity and board governance variables. Furthermore they emphasize the concerns reverse causality could cause. Director compensation structure and firm performance could affect both the incentive of women to join a firm and the incentive of a firm to hire women. Adams and Ferreira (2009) address these problems by IV methods, in which they use an instrument for the fraction of female directors.
2.2.1.2 Empirical Results
Adams and Ferreira (2009) conducted several regressions in their investigation. The most relevant one is the regression between the share of female directors and performance. They tested whether performance was influenced or not by the share of female directors. They used a multiple regression model which included the fraction of females on the board and some control variables, such as board size; board independence; log(sales); the number of business segments; year dummies; and industry dummies. To measure performance they used two measures, being the dependent variables in their regressions. First of all they used tobin's q. They define tobin's q as the ratio of the firm’s market value to its book value of assets. Many former studies used tobin's q as well and therefore it seems to be a valid measurement instrument for performance (Adams and Ferreira, 2009). Tobin's q is a market‐based measure for performance. Secondly, they use an accounting‐based measure for performance, the return on total assets (ROA), as a measure for performance. The return on total assets is defined as net income divided by total assets.
To account for omitted variable problems Adams and Ferreira (2009) add firm‐fixed effects. To account for problems caused by reverse causality, they came up with an instrument that was not used in other empirical studies investigating the relation between the share of female directors on the board of directors and firm performance. The instrument they use is the fraction of male directors on the board of directors who sit on other boards on which there are female directors.
Adams and Ferreira (2009) argue that the greater this fraction is, the greater the gender diversity on the board should be (Adams and Ferreira, 2009). This instrument, they continue, is correlated with the share of female directors on the board of directors, but definitely not with the performance measures used, except for correlations through the control variables used. Therefore this makes a good instrument for preventing possible reverse causality problems, and to sort out the differences in influence on performance by the shares of male and female directors.
Adams and Ferreira (2009) show, according to them, different results than prior studies. For example Shrader et al (1997), who found that the relationship between the share of females on the board of directors and firm performance is positive for US data, especially when accounting‐based measures are used (Shrader et al, 1997). Adams and Ferreira (2009) show that, for both tobin's q and ROA, the relation between the share of female directors and firm performance is negative. They argue that a more diverse board in terms of gender is harder to monitor, and therefore decreases performance. This is consistent with the fact that over monitoring could decrease value (Adams and Ferreira, 2009). However, they show that gender diverse boards appear to be valuable when governance within in the firm is weak.
Besides the regression between share of female directors and performance, Adams and Ferreira (2009) conducted some other regressions. First of all Adams and Ferreira (2009) wanted to show differences in attendance between males and females. To test this they estimated a probit model in which the independent variable, the attendance of at least 75% of board and committee meetings, is one if the director did not meet the 75% attendance in a given year and a zero if the director did meet the 75% attendance. The results showed that almost all directors in their sample meet the attendance requirement, only 2.38% did not meet the 75% attendance. Adams and Ferreira (2009) argue that this could be due to the fact that directors do not want to be known as directors with attendance problems. After this regression they continued by splitting the directors into two groups, one consisting of inside directors and one consisting of outside directors. They did this because they assumed the reasons for not meeting the 75% attendance to be different for these groups. Outside directors might have other jobs for which they have to be present at certain moments (Adams and Ferreira, 2009). This might force them to choose and therefore miss meetings. In this regression Adams and Ferreira (2009) tested whether there are differences in attendance for males and females, both being inside directors. The results from these regressions show a negative coefficient on females (the female dummy is negative) which means females have less of an attendance problem than males. Even after controlling for director characteristics, such as gender or age, they find that female directors behave better than male directors. However, Adams and Ferreira (2009) find that the presence of women on the board positively affects the behavior of men. Their attendance increases when a board is more diversified.
2.2.1.3 Conclusions
For the purpose of this paper, the most relevant findings of Adams and Ferreira (2009) can be summarized as follows. First of all Adams and Ferreira (2009) do not find a positive relationship between the share of female directors and firm performance. If anything, it appears to be negative. This negative relationship holds for both tobin's q and ROA. On the other hand they find that making a board of directors more diverse positively affects the behavior of male directors. Their attendance on board meetings increases (Adams and Ferreira, 2009). 2.2.2 The Impact of Gender Diversity on the Performance of Business Teams The article, written by Hoogendoorn, Oosterbeek and Van Praag (2013) investigates gender diversity using a field experiment. The main reason they did this investigation is because the evidence on causal effects was too thin. I will describe the data and methodology in section 2.2.2.1, which includes the set‐up of the experiment. In section 2.2.2.2 I will analyze the results of this field experiment.
2.2.2.1 Set‐up, Data and Methodology
In collaboration with the Junior Achievement Young Enterprise Start‐Up Program, which is the leading entrepreneurship education program in both the United States and Europe, Hoogendoorn et al (2013) conducted the field experiment with first‐year students from the Amsterdam College of Applied Sciences. The students were assigned randomly to a group, which makes this experiment different from other empirical studies on this topic. The groups consisted of approximately twelve students, both males and females. The program was called the Start and Run Business in Entrepreneurship Program, and was compulsory for the students' first year. The program lasted the entire academic year in which students had to set up a small‐sized company. The students had to deal with many firm‐related issues such as selling stock, electing officers, dividing tasks, and many more. As said, teams consisted of twelve students, which had to be divided into a managing group and a ''normal'' employees group. After half a year the groups would switch, giving everyone the opportunity to have a leading function, thereby influencing the performance more.
The data used for this experiment was obtained from students participating in the experiment. 550 students started the year, and were randomly assigned to one out of forty‐five groups, though single‐sex groups or groups with one male or female were not allowed. The results of the groups showed that no group existed with less than 17% or more than 58% females. During the academic year 104 students dropped out, which changed the average number of group members from approximately twelve to approximately ten students. However the dropouts did not affect the shares of males and females.
The most important regression Hoogendoorn et al (2013) estimated is the regression on the share of females in the groups, with performance as the dependent variable. To measure performance, Hoogendoorn et al (2013) used sales and profit.
2.2.2.2 Empirical Results
The field experiment conducted by Hoogendoorn et al (2013) results in the following. First of all they find an inverse U‐shape in the relation between the share of females on a team and team performance, measured by sales and profits. The inverse U‐shape indicates that any share of females between 0.2 and 0.5 (20‐50% females) increases team performance, measured in sales. However, when the share of females exceeds 0.5 the sales weaken as the share of females increases. For the other performance measure, profits, they find approximately the same pattern. As long as the share of females is below 0.5 the profits increase. For any share higher than 0.5 the profit function is flat, indicating profits do not increase, neither do they decrease. On top the graphs show that the optimal mix of males and females is an equal mix. For any other mix performance decreases, although they mention the fact that they were not able to examine the effects for shares of females exceeding 0.6. They assume that the more dominated a group is by women, the worse the performance will be though. On the other hand, a group dominated by males, has a weaker performance for sure, as indicated by the inverse U‐shape.
2.2.2.3 Conclusions and Relevance
This investigation is very relevant for my own investigation. Especially the part in which they use a continuous variable on females. Hoogendoorn et al (2013) find an inverse U‐shape showing that for certain shares of females the performance increases. For other shares however, the performance decreases.
2.2.3 Link to other evidence
In the previous two sections I discussed the most relevant papers for my investigation. One of them, Adams and Ferreira (2009) finds evidence showing that a more diverse board performs worse. The other, by Hoogendoorn et al (2013) finds the exact opposite. In this section I will provide evidence from other papers that investigated this relation.
There are some papers that found no relationship at all between the share of female directors and firm performance. First of all there is the article by Ayuso et al (2007) that investigated the stakeholders' approach towards corporate governance. They find that a diversified board in terms of gender does not improve, nor worsen performance. Randøy et al (2006) investigated the 500 largest firms in Scandinavia (Sweden, Norway and Denmark) and found no significant evidence
showing any effect of the share of female directors on firm performance, using return on total assets (ROA) and the performance of the stock. Farrel and Hersch (2005), using US data, found that the market's reaction is insignificant in terms of abnormal returns when companies announce the addition of a female to the board of directors (Farrel and Hersch, 2005). Furthermore Rose (2007), using tobin's q as the performance measure, found no relationship between a more gender diverse board and firm performance within Danish listed firms. Lastly, Smith et al (2006) do not find significant evidence showing the relationship, using US data.
Few papers find evidence for an improved performance when boards are more diverse in terms of gender. Carter et al (2003) find a positive relationship using tobin's q as the measure for performance. Erhardt et al (2003) find positive relations as well using more accounting based measures, such as return on equity (ROE) and return on total assets (ROA). This is confirmed by the results Shrader et al (1997) found, saying that in the US the more accounting‐based measures tend to give positive results for the relationship between the share of female directors and firm performance. Campbell and Minguez‐Vera (2008) found a positive relationship in Spain. The fact that they found a relationship at all, according to them, is extraordinary, since Spain historically had few women on the board of directors (Campbell and Minguez‐Vera, 2008). Francoeur et al (2008) investigated the relationship in Canada, using the 500 largest Canadian firms. They found that in complex environments, a higher share of females on the board of directors positively influences abnormal returns (Francoeur et al, 2008). A negative relationship between gender diverse boards and firm performance was found by Adams and Ferreira (2009) in the United States. In Scandinavia, Bøhren and Strøm (2007), Ahren and Dittmar (2012) and Daunfeldt and Rudholm (2012) found this negative relationship as well. Daunfeldt and Rudholm (2012) included non‐listed companies in their investigation, which made it impossible for them to use tobin's q, market‐to‐book ratio or abnormal returns. They therefore used the return on total assets (ROA). Ahren and Dittmar (2012) used tobin's q because they argue that, due to changes in accounting rules, other measures are not reliable.
2.3 Hypothesis
The conclusion on the relationship between the share of female directors and firm performance is not clear. A lot of studies show different results. These can be due to differences in time periods, geographical location or methods used by the various studies (Campbell and Minguez‐Vera, 2008). What I can conclude so far is that there are more arguments in favor of a diversified board in terms of gender (section 2.1.1) than there are against (section 2.1.2). However, following the article by Adams and Ferreira (2009), since in my investigation I will use US data, my hypothesis will be the following. If a relationship between the share of female directors and firm performance is shown, I expect it to be
a negative relationship. Moreover, I expect to show a U‐shaped graph indicating this relationship. The graph will indicate the fact that the most gender diverse board, having equal amounts of males and females on it, is the minimum point on the graph. 3. Data and Methodology In this section the data and methodology will be discussed. In section 3.1 I will describe my data. In section 3.2 the methodology used will be discussed. 3.1 Data To test my hypothesis I will use a sample of American firms, being the Standard and Poor's (S&P) 500 firms. I chose this sample because it is well‐known and gives me the opportunity to test my research question along the 500 biggest firms in the United States. I am aware of the fact that the results this sample will provide might not be representative for smaller firms, still I think it will give a good view on how performance is affected (or not) by the diversification of boards in terms of gender. I will use data from 2008‐2012 to get a broad view. By using multiple years, I prevent the data to be extreme. By extreme I mean an extremely good or extremely bad year. If either of these two is the case when using one year only, the results obtained will not be very reliable. To prevent this from happening, I use five years. The data set I will use was obtained from Wharton research data services (wrds). To do my investigation I will need data on directors, such as gender and age, as well as data on the performance of the firm. I will use three performance measures in order to check my research question from different point of views. I will first of all use the growth in sales, since growth in sales represents one of the most fundamental goals of businesses, while also being an accurate and available measure of performance (Boone et al, 2007). Besides I will use two commonly used accounting‐based performance measures, being the return on total assets (ROA) and the return on equity (ROE). Return on total assets is defined as the net income divided by total assets. Return on equity is defined as net income divided by equity. Both of these measures are used by Daunfeldt and Rudholm (2012). They used specifically these measures to be able to compare listed firms with non‐ listed firms. By using the ROA and ROE, it might be able to compare my results with results from other investigations that look at non‐listed companies in the United States. Furthermore, Shrader et al (1997) argue that accounting based‐measures tend to give positive results in regressions between the share of female directors and firm performance in the United States. Therefore, by using these two performance measures, I can show whether this is still the case, or not.
3.2 Methodology To test my hypothesis I will use a regression model, in which performance is the dependent variable and the most important independent variable is the continuous variable on the share of females on the board of directors. Furthermore I will add experience in the form of age and board size, to come to the following model:
Performance Measure = ß0 + ß1*(share of females) + ß2*(share of females^2) + ß3*(age) + ß4*(age^2) + ß5*(Board Size)
As mentioned in section 3.1, three different performance measures will be used to conduct the regressions. These three are the growth in sales, the return on total assets (ROA), and the return on equity (ROE).
The first two independent variables are the share of females on the board of directors, and the squared form of the share of female directors. Combined, these two are the most important variables in my regression. The squared form of the share of female directors will make it possible to account for non‐linear relationships. Hoogendoorn et al (2013) did this as well and they were able to show an inverse U‐shape, after concluding the relationship was non‐linear for both of their performance measures used. Hoogendoorn et al (2013) showed that increasing the share of females on a group positively influences the performance of a group up to a certain share. Based upon my hypothesis however, I expect differently. I expect the sign of ß1 to be negative, while the sign of ß2 is expected to be positive. The results then will be exactly opposite to the results found by Hoogendoorn et al (2013) for these two variables. This in turn will show the exact opposite figure, a normal U‐shape. Of course, the coefficient ß2 has to be significant to show the non‐linear relationship in the first place. If not, there is a linear relationship, and I will not be able to show a U‐ shaped curve.
Besides the two variables for the share of female directors I will use three more control variables. I will use age and age squared, because age is an indication of experience. I use Age‐
Squared because with time, experience does not contribute as much as it used to. The marginal
contribution of experience can be seen as an inverse U‐shape, having a maximum point somewhere in the middle of all working years. To show this effect I use the squared version of age. Moreover, the signs of the coefficients for Age and Age‐Squared have to be correct and significant. The sign for the coefficient of Age has to be positive, while the sign for the coefficient of Age‐Squared has to be negative. If both are significant, an inverse U‐shape can be shown for the experience. This is in accordance with the literature (Daunfeldt and Rudholm, 2012; Levi et al, 2008). I follow Adams and
Ferreira (2009), who used age as an indicator for experience as well. Because I will test my regression using boards of directors I will have to average the age of all directors on the board of one firm to get the average age of the boards.
The last variable I add is a variable to control for board size. Board size is defined as the total number of directors on a board. Most of the former studies investigating the relationship between the share of female directors and firm performance use this variable to see the efficiency of a board of directors (Adams and Ferreira, 2009). Daunfeldt and Ruholm (2012), Ahren and Dittmar (2012) and Campbell and Minguez‐Vera (2008) use board size as a control variable as well. Based upon prior literature I expect the sign of the coefficient for Boardsize to be negative (Adams and Ferreira, 2009; Daunfeldt and Rudholm, 2012; Ahren and Dittmar, 2012).
All the data I use will contain information about a company for five years. To be able to perform regression analysis I will average all numbers. The performance measures will in this case be the average performance over the five year period. The independent variables I use will also be averaged. Section 4 will explain this in more detail.
3.3 Checked Data
After checking all the data I had to remove twenty‐five observations due to various reasons. First of all six firms were removed because at least one director's age was unknown. Furthermore I removed four firms because the gender of the directors was unknown. Lastly, another fifteen firms were removed because they had unrealistic (e.g. ROA of 3232.4%) or no result for the performance measures. This leaves me with 475 observations of boards to do the investigation with. 4. Results In this section I will show the results from the regressions performed. In section 4.1 I will show the descriptive statistics on both the dependent and independent variables. In section 4.2 I will provide the empirical results that followed from the regressions I performed. 4.1 Descriptive Statistics As mentioned in section 3.1, the original sample of 500 firms is reduced to 475 firms due to various reasons. All data on these 475 firms is available and will be used. The 475 firms together had 5176 directors in the time period 2008‐2012. 4242, or 82%, of the directors were males, and 934, or 18%, of the directors were females (Table 1). When looking at males and females separately, I can see that in my sample the males are, on average, around 2 years older than females (Table 2). Furthermore
the females provide the youngest director, 31 years of age, while the males provide the oldest director, 96 years of age (Table 2). Table 1 (Total Estimations): Total Percentage of Total No. of Directors 5176 (49.65) 100% No. of Males 4242 (46.59) 82% No. of Females 934 (24.78) 18% Notes: Between brackets is given the standard deviation Table 2 (Directors' ages):
Observations Mean Minimum Maximum
Females 934 61.5 (2.13) 31 79 Males 4242 63.7 (3.79) 36 96 Notes: All numbers, except the number of observations, are years. Between brackets is given the standard deviation.
Next I grouped all the directors belonging to the same firm, giving me 475 different boards. I took into account the fact that board seats were switched from one director to another during these years. This however, did not affect the total board size of a single company. When calculating the share of female directors however, the numbers could change. Therefore, when calculating the share of females, I took into account the fact that when a board seat was switched from either male to female or female to male, this affected the average share of female directors on those boards. To prevent wrong shares enter the calculation I did to get the average share in Table 3 I first calculated the average share of females for the boards on which a situation like that occurred, before letting these numbers enter the calculation of the average share of female directors. To make clear what I did exactly, I provide an example. Assume a board with ten directors in 2008 (the first year of my sample), three females and seven males. After two years, at the beginning of 2010, a male director is replaced by a female director. For the average share of females, the following happened. For two years, three females were on the board ((2/5)*3=1.2) and for three years, four females were on the board
((3/5)*4=2.4). This means the total number of female directors on this board during the five‐year time‐period used, was 3.6. The average share of female directors for this board was 0.36, because the total number of directors remained ten over the entire period. Situations like this occurred rarely in my sample, but when they did occur I used this method to overcome the problems. Of course, when a male was replaced by a male, or a female by a female, nothing changed in terms of the share of female directors. After I had found all the average shares of female directors per board, I had 475 observations, which are shown in Table 3. Of course a similar situation occurred with the average age of a board when a board seat was switched. I used the same method to calculate the average age of the board first, before using this ''single'' observation to calculate the overall average age in Table 3. Table 3 provides the descriptive statistics about the boards. On average a board consists of 10.7 directors, which is considerably lower than the maximum, a board consisting of 34 members. I chose to keep this outlier because in my regressions the share of female directors is important, not the total amount of directors. Even the fact that this board had only two female directors, or a share of female directors of 5.9%, it still contributes since the lowest share in my sample is 0.0. The average age of a board is approximately 62 years, while the youngest and oldest board are 53.3 and 71.6 years respectively. Very important for my investigation is the share of female directors. Hoogendoorn et al (2013) did not find a share higher than 0.58 in their field experiment in the Netherlands. The highest share in my regression will be 0.56, while the lowest is 0.0. On average the share of females on a board of directors is 0.17.
Table 3 (Board information):
Variable Mean Minimum Maximum
Average Age 62.32 (3.01) 53.3 71.6 No. of Directors on Board 10.69 (2.28) 5 34 No. of Females on Board 1.86 (1.14) 0 7 No. of Males on Board 8.83 (2.14) 3 32 Share of Females 0.17 (0.10) 0 0 .56 Notes: Average Age means average age per board, the mean represents the mean age of all boards. Share of Females represents the average share of females per board, the mean represents the mean share of all boards. Between brackets is given the standard deviation.
When paying extra attention to the number of directors, one can easily see a difference between the total number of directors (5176) provided in Table 1, and the mean number of directors (10.69) provided in Table 3 multiplied by the total number of boards (475). This gives a total of 5078 directors. This difference occurred because board seats changed during the five‐year time‐period. Table 1 provides information about all directors who were on one of the 475 boards in the time period used in this sample, while Table 3 provides information on the boards themselves. One observation in Table 3 implies an average per firm. As I mentioned earlier, the fact that a board seat changed from one director to another, did not affect the total number of directors. The performance measures I will use in my regression are the growth in sales, the return on total assets (ROA), and the return on equity (ROE). I will look at the averages of these variables over the five‐year period (2008‐2012). In total there are 2375 (475*5) observations for each performance measure. Appendix 1 shows the 2375 single observations for the performance measures. I averaged all performance measures for all firms. Then I created Table 4, showing 475 observations, one for each firm in my sample. A single observation in Table 4 therefore implies the average of five observations. From Table 4 I can see that on average, the firms had a growth in sales of almost 10%, while the means of the return on equity and return on total assets are 16.9% and 6% respectively. Looking at the minimums and maximums it is clear that there are large differences for the return on equity, ranging from ‐61 up to 396. This means there are outliers, which could influence my results. I cannot remove these outliers however, since they might be true values and have to be taken into account. For the growth in sales the range is smaller already (‐27 up to 72), but the range for the return on assets is smallest (‐24 up to 36), which will probably give the most reliable results when regressed.
Table 4 (Performance Measures): Performance
Measures Observations Mean Minimum Maximum
Growth in Sales 475 9.61 (4.56) ‐26.80 72.15 Return on Total Assets (ROA) 475 6.04 (2.30) ‐23.72 36.06 Return on Equity (ROE) 475 16.91 (26.04) ‐61.04 396.33 Notes: All performance measures are measures over a five‐year time period (2008‐2012), this means every observation in this table indicates an average already. Means, standard deviations, minimums and maximums are percentages. Between brackets is given the standard deviation.
4.2 Empirical Results
In this section I will provide the empirical results that followed from the regressions I performed. To do my regressions I made a model, explained in section 3.2. To test the model for the three performance measures I performed OLS (ordinary least squares) regressions. For every performance measure I did the same. I started by regressing the variable for the share of female directors on the dependent variables first. Then in the second regression I added the three control variables Age,
Age‐Squared and Boardsize. Lastly, in the third regression I added the share of female directors
squared (ShareFem‐Squared) as well. I did this to be able to see whether there is a linear or non‐ linear relationship. When there is a non‐linear relationship the coefficient on the variable for the squared version of the share of female directors should be significant. There needs to be a non‐linear relationship in order to show a U‐shaped (or inverse U‐shaped) graph. To show a U‐shape, indicating that the more diverse a board of directors is in terms of gender, the worse will performance be, the sign of the coefficient on the share of female directors needs to be negative (and significant), while the sign of the coefficient for the squared form of the share of female directors should be positive (and significant). Only then I will be able to show a U‐shape. When the coefficient for ShareFem‐
Squared is not significant, but the coefficient for ShareFem is, a linear relationship is shown. When
both are insignificant no relationship is shown at all. The next three sections will show the results obtained from the regressions on the three performance measures separately. Section 4.3 will show additional results, omitted variable bias and reverse causality problems. Section 5 will discuss the results shown in the following sections
4.2.1 Growth in Sales
The three different regressions on this dependent variable resulted the following. First of all, model one (Table 5), showing a regression between ShareFem (independent variable) and the growth in sales (dependent variable), shows that the coefficient for ShareFem is negative (‐19.64) and very significant (using a one percent significance level). The adjusted R‐squared belonging to this regression shows that by itself, the share of female directors explains almost 5% percent of the growth in sales, while the F‐statistic (9.22) shows the model altogether is significant at 1%.
When adding three control variables (Age, Age‐Squared and Boardsize) in model two, I can see that the coefficient for ShareFem remains very significant (at 1%), although the coefficient itself changes slightly, from ‐19.64 to ‐18.75. The three variables added are all significant using a 1% significance level. The coefficient for Age is negative (‐14.58), while the coefficient for Age‐Squared is positive (0.11). Therefore, the total contribution of experience (indicated by Age and Age‐Squared) is negative. This result by itself is strange, since the signs for both Age and Age‐Squared are exactly opposite than expected. I will elaborate on this contradictory result when discussing the results in
section 5. Boardsize lastly, has a negative (‐0.70) coefficient, while also being significant using a 1% significance level. Model two altogether is significant at 1%, following the F‐statistic (5.77). This model furthermore explains better than model one the growth in sales, as indicated by the adjusted R‐squared. Model two explains almost 9%.
To see whether the relationship between the share of female directors and firm performance (growth in sales in this regression) is linear or not I added ShareFem‐Squared in the third model. In order to show a quadratic relationship, the coefficient for this variable should be significant. Table 5 shows that in model three the coefficient for ShareFem‐Squared is not significant. Therefore, the relationship between the share of female directors and the growth in sales is linear. Graphically this is shown in appendix 2. From the figure in the appendix it follows that the growth in sales decreases when the share of female directors increases. This follows from Table 5 as well. The coefficient for
ShareFem is negative (‐26.45) and significant at 5%. The coefficient for ShareFem‐Squared is positive
(18.03) but not significantly different from zero. In total, the contribution to the growth in sales by the share of female directors is negative. As said, this follows from both Table 5 and appendix 2. When looking at the other variables in model three, I can see that the coefficients remained almost equal, while they all became less significant than in model two. Age first of all is still negative (‐14.62) but instead of being significant at 1% the coefficient is now only significant at 5%. The coefficient for Age‐Squared (0.11) remained the same, while again the significance decreased from 1% to 5%. Lastly, the coefficient for Boardsize increased from ‐0.70 in model two to ‐0.67 in model three, while again the significance decreased from 1% to 5%. The fact that the coefficients became less significant when switching from model two to model three is also indicated by the F‐statistic, which decreased from 5.77 in model two to 4.65 in model three. This is however, still significant at 1%, which means model three altogether is good. The adjusted R‐squared confirms this. Model three explains 12% of the growth in sales, while model two only explained 8.8%. Concluding I can say that while model three best explains (highest R‐Squared) the growth in sales, it did not show a non‐linear relationship. Furthermore, the results shown for the variables Age and Age‐Squared seem odd. This will be discussed in section 5 in more detail.
Table 5 (Regressions on Sales Growth): (1) (2) (3) ShareFem ‐19.64*** (6.47) ‐18.75*** (6.69) ‐26.45** (17.74) ShareFem‐Squared 18.03 (39.99) Age ‐14.58*** (5.79) ‐14.62** (5.87) Age‐Squared 0.11*** (0.046) 0.11** (0.047) Boardsize ‐0.70*** (0.29) ‐0.67** (0.30) Constant 12.99*** (1.29) 481.86*** (181.73) 471.01*** (184.08) F‐Statistic 9.22*** 5.77*** 4.65*** R‐Squared 0.059 0.097 0.14 Adjusted R‐Squared 0.049 0.088 0.12 Notes: Between brackets is given the standard deviation Model 1 included the variable ShareFem only Model 2 included all variables except for ShareFem‐Squared Model 3 included all variables * means significant at 10% ** means significant at 5% *** means significant at 1% 4.2.2 Return on Total Assets (ROA) The results for the regressions on the return on total assets are shown in Table 6. Model one first of all shows the regression between the share of female directors and the return on total assets only. From this regression I can see that the coefficient for ShareFem is negative (‐4.44), but not significant. The expected consequence is that the model altogether is not significant, which is confirmed by the non‐significant F‐statistic (2.49). The adjusted R‐squared lastly, shows that this model explains less than 1% of the return on total assets.
Model two includes, besides ShareFem, also three other control variables (Age, Age‐Squared and Boardsize). In this model, the coefficient for ShareFem is still not significant. The coefficient is still negative (‐4.05) though. Experience, indicated by Age and Age‐Squared, showed no significant coefficients in model two. Age is negative (‐2.20), while Age‐Squared is positive (0.02). The only significant variable in this model is Boardsize. The coefficient is negative (‐0.34) and it is significant at 5%. In total, model 2 explains 2% (adjusted R‐squared) of the return on total assets, while the model altogether is significant at 1% (F‐statistic is 2.84).
Important for my investigation is to show a non‐linear relationship between the share of female directors and performance. To do so, model three was formed. In this regression I added
ShareFem‐Squared, which has to be significant in order to show non‐linearity. Moreover, it has to be
positive to show a U‐shape (while ShareFem has to be negative and significant). In the regression on the return on total assets, the coefficient for ShareFem‐Squared is positive (14.53) and significant (at 5%), and therefore the relationship between the share of female directors and the return on assets has a non‐linear character. Because the coefficient for ShareFem is negative (‐6.00) and significant (at 5%), which is the second requirement in order to show a U‐shape, this relationship will show a U‐ shape. Figure 1 shows this. From the figure it can be seen that, starting from a zero share of female directors, the performance decreases when the share of female directors increases. However, this graph does not show that the more diverse a board of directors becomes in terms of gender, the worse performance will be. The minimum in the graph is 0.21, which means that for a share of female directors of 0.21 the performance will be weakest (measured along the ROA). From 0.21, the graph, and thereby performance (ROA), increases when the share of female directors increases. This means that up from a share of female directors of 0.21, the performance increases as a board of directors becomes more diverse. Up to the most diverse board in terms of gender (share of female directors of 0.50). Furthermore, it can be seen that from a share of female directors of 0.21, the return on total assets keeps increasing up to the highest share (0.56) in my sample. Although I was not able to look at higher shares of female directors, I assume the graph to continue this path. Therefore, the higher the share of female directors becomes, the better will the performance (ROA) be. The most diverse board in terms of gender is a board with a share of female directors of 0.50. When looking at Figure 1, it can be seen easily that for a share female directors of 0.50 the return on total assets is higher than the return on assets is when the share of female directors is zero. Therefore, this model is not consistent with my hypothesis stating that when there is a relationship between the share of female directors and performance, this relationship is negative. Model three shows a positive relationship between the share of female directors and performance for any share larger than 0.21. The positive relationship follows from Table 6 as well. ShareFem is negative (‐6.00), while ShareFem‐squared is positive (14.53). They are both significant using a 5% significance level. In total, the contribution of the share of female directors is positive. This is consistent with the graph in Figure 1.
Looking at the other three variables in this model I can see that they are all significant. The coefficient for Age is negative (‐2.11) and significant at 5%, while the coefficient for Age‐Squared is positive (0.02) and significant at 10%. Again these results are somewhat strange, but as said in the former section I will discuss this in section 5. In total, the influence of experience, indicated by Age and Age‐Squared, is negative. Boardsize is the last variable I use in my regression. The coefficient is negative (‐0.33) and significantly different from zero using a five percent significance level. Model
three altogether is significant (F‐statistic of 4.77) and it explains 9% (adjusted R‐Squared) of the return on total assets.
Concluding I can say that model three partly showed the expected results. The sign of the coefficient of ShareFem was negative while the coefficient of ShareFem‐Squared was positive. This showed a U‐shape consistent with my hypothesis, and contrary to the shape Hoogendoorn et al (2013) showed. However, since the coefficient for ShareFem‐Squared is larger in an absolute way than the coefficient of ShareFem, the total contribution of the share of female directors on the return on total assets is positive. Graphically this showed that the most diverse board (share of female directors of 0.50) did not have the worst performance, as my hypothesis stated. Therefore, while I showed the shape I wanted to show, the results do not satisfy my hypothesis. Table 6 (Regressions on ROA):
(1) (2) (3) ShareFem ‐4.44 (2.81) ‐4.05 (2.92) ‐6.00** (7.77) ShareFem‐Squared 14.53** (17.53) Age ‐2.20 (2.53) ‐2.11** (2.57) Age‐Squared 0.02 (0.02) 0.02* (0.02) Boardsize ‐0.34** (0.13) ‐0.33** (0.13) Constant 6.80*** (0.56) 81.11 (79.53) 82.08 (80.68) F‐Statistic 2.49 2.84*** 4.77*** R‐Squared 0.005 0.03 0.11 Adjusted R‐Squared 0.003 0.02 0.09 Notes: Between brackets is given the standard deviation Model 1 included the variable ShareFem only Model 2 included all variables except for ShareFem‐Squared Model 3 included all variables * means significant at 10% ** means significant at 5% *** means significant at 1%
Figure 1 (Return on Assets and Share of Female Directors): Notes: X‐axis: Share of Female Directors Y‐axis: Return on Total Assets Fitted values means only the important values are shown. It starts with a ROA of 5, because no share of female directors belonged to a lower value of ROA. On the other hand, no values larger than 8.5 are needed to show this graph, because no share of female directors in my sample belongs to a value larger than 8.5. 4.2.3 Return on Equity (ROE) The last performance measure I used to show the relationship between the share of female directors on the board of directors and firm performance is the return on equity (ROE). The results are shown in Table 7. Again I started with a model regressing the share of female directors on the performance measure, return on equity in this case, only. Model one shows that the coefficient for ShareFem is positive (12.79), but not significant. This in turn causes the model altogether to be insignificant, as indicated by the F‐statistic (1.20). The adjusted R‐Squared furthermore shows that approximately zero percent of the return on equity is explained by this model.
Model two is again model one enhanced with three extra variables. Age, Age‐Squared and
Boardsize were added to the model and this regression showed the following. ShareFem remains
positive (15.36), but is still not significant. Experience, indicated by Age and Age‐Squared, is not proven to have an influence on performance (measured by return on equity), since Age and Age‐
Squared are both not significantly different from zero. The signs of the coefficients for these two
variables are exactly opposite to the results shown in sections 4.2.1 and 4.2.2. The coefficient for Age is positive (14.36) in this regression, while the coefficient for Age‐Squared is negative (‐0.11). This is however, consistent with the expectations of the signs for these variables. The last variable used is
5 6 7 8 F it ted v a lu es 0 .2 .4 .6 ShareFem2
Boardsize, which has a positive (0.14) coefficient in this regression. Furthermore, it is the only
variable in this model that is significant, though it is significant at 10% only. Following the fact that only one variable is significant, this model is not significant altogether (F‐statistic of 1.08). Lastly, model two explains less than 1% (adjusted R‐Squared) of the return on equity.
Model three again includes the variable ShareFem‐Squared, which, when significant, shows the model is non‐linear. However, following from Table 7, it can be seen that the coefficient for
ShareFem‐Squared is not significant. Therefore, this model is assumed to be linear. However, since ShareFem is not significant, a linear relationship is not proven either, and therefore cannot be shown
graphically. The regression on the return on equity therefore shows no relationship at all between the share of female directors and firm performance (ROE). When looking at model three, I can see that only two variables (Age‐Squared and Boardsize) are significantly different from zero. However, they are significant at 10% only. As said both ShareFem and ShareFem‐Squared are not significant. Their coefficients are 34.57 and ‐46.15 respectively, which are both exactly opposite signs when compared with the regressions performed in sections 4.2.1 and 4.2.2.
Age and Age‐Squared show almost the same coefficients in model three, as they showed in
model two. The coefficient for Age is positive (13.56), while the coefficient for Age‐Squared is negative (‐0.10). The only difference with model two is that in model three Age‐Squared is significant at 10%. The last variable used in this regression is Boardsize, which is positive (0.69) and significant at 10%.
Altogether model three is not good at predicting the return on equity, following the insignificant F‐statistic (1.17). This is confirmed by the adjusted R‐Squared, which shows that the model explains, in total, 0.3% of the return on equity.
Concluding I can say that the results shown in these three regressions are not consistent with my hypothesis. The signs of ShareFem and ShareFem‐Squared are both exactly opposite to what they should be in order to show a U‐shape. However, even if the signs were correct, since both of the variables are not significant, they do not prove any relationship at all. Therefore, these regressions showed a negative answer to my research question. No relationship between the share of female directors and firm performance, measured using the return on equity, is shown.