CEO GENDER COMPENSATION GAP
Can we speak of a gender wage gap between male and
female CEO’s looking at S&P500-Indexed firms?
Abstract:
The existence of a gender wage gap between male and female CEO’s might be problematic in terms of reducing the motivation to perform of a female CEO. Therefore, the aim of this study is to investigate whether we can speak of a gender wage gap between male and female CEO’s looking at S&P500-indexed firms in the year 2016. We approach this study empirically using linear regression. The variable of interest is the gender of the CEO and furthermore, we control for age, tenure, performance and whether the firms headquarter is located in a state requiring equal pay law legislation. Results show that being a female CEO has a positive effect on the total compensation, counterintuitive to what one may expect. Despite this effect being statistically insignificant, we can conclude that we cannot speak of a gender wage gap between female and male CEO’s looking at S&P500-indexed firm in 2016.
Name: Demi Valentijn Track: Financiering en Organisatie Student number: 10982760 Supervisor: Konstantinos Ioannidis Programme: Economie en Bedrijfskunde Date: January 6th 2018
Statement of Originality
This document is written by Demi Valentijn who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are 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.
Table of Contents
CHAPTER 1: INTRODUCTION……….. 5
CHAPTER 2: RELATED LITERATURE………... 7
2.1 CEO compensation gap………...………… 7
2.2 Firm performance...……….. 8
2.3 Risk preferences……….. 9
2.4 Other explanations…...………10
CHAPTER 3: RESEARCH METHODOLOGY………...………..12
3.1 Research Model………...……….. …….12 3.1.1 Statistical hypotheses...…………...……….12 3.1.2 Functional form………....13 3.2 Data collection………...……….13 3.3 Research variables………...14 3.3.1 Dependent variable………...14 3.3.2 Independent variables………...14 3.4 Research method………...………..16 CHAPTER 4: RESULTS…………...………17
4.1 Validity of the multiple linear regression assumptions……….………...17
4.1.1 No outliers………...17 4.1.2 Linearity………..17 4.1.3 No multicollinearity………18 4.1.4 Normality……..………..19 4.1.5 Homoscedasticity………19 4.1.6 No autocorrelation………..20 4.2 Descriptive Statistics………..21
4.3 Regression and discussion ………..…...22
CHAPTER 5: CONCLUSION, DISCUSSION & SUGGESTIONS FOR FURTHER RESEARCH ………...………25
5.1 Conclusion………...25
5.2 Discussion………...25
5.3 Suggestions for further research………...26
REFERENCES………....27
APPENDICES ………...30
Appendix 1………....28
Appendix 2………29
Chapter 1: Introduction
The gender wage gap is defined as the difference between male and female median wages divided by the male median wages (OECD, 2017). According to GlassCeiling (2015), men’s wages rise until they are 50 to 55 years old, with a median salary of $75,000, while women’s salaries stop rising when they are 35 to 40 and have a median of $49,000 (see Appendix 1). This is problematic because women still earn less than men despite equal pay law (The Economist Group Limited, 2017). In 2016, full-time working women in the United States were paid just 80 percent of what men were paid, indicating a gender wage gap of 20 percent (American Association of University Women, 2017). The gender wage gap could be a reducing factor of the motivation of female employees, managers, executive officers and most importantly, the CEO, Chief Executive Officer.
Searching through academic journals and other literature, it is made clear that CEO compensation differing between genders is a very popular research topic. Loads of researchers have devoted their time looking for the evidence necessary to establish the existence of the gender wage gap. There are some previous studies who claim that the gender wage gap of CEO’s is real (Bertrand & Hallock, 2001, p.17). Besides that, there are also studies who have showed the CEO gender wage gap to be nonexistent (Bugeja, Matolcsy & Spiropoulos, 2012, p.16). Because of this ambiguity, it is most important to take the topic seriously and to perform more research on it. The gender wage gap resembles discrimination. Closing the gap will help create justice and equal chances in society. Because of this inequality, it is, as mentioned earlier, likely to reduce the motivation of females. By closing the gap, performance will be improved and especially for CEO’s it is very important to realize a sufficient overall performance of the firm.
The job as a CEO is the most important position in a firm. It is a job that only CEO’s can do because everybody in the organization is focused much more narrowly, and most important, in one direction (Lafley, 2009). A CEO must focus on every direction of the firm and sees responsibilities that no one else can see. On top of that, a CEO is the only one held accountable for the performance and results of the company (Lafley, 2009). Because of this, it is relevant to examine whether the gender of the CEO has an effect on his or her total compensation. Therefore, my aim is to research whether we can speak of a gender wage gap between male and female CEO’s looking at S&P500-Indexed firms.
Inspired by recent related literature, for example Newton & Simutin (2012) where to some extent a similar dataset is used, S&P500-Indexed firms will be used, because these firms have similar firm sizes and are located in the United States. The reason for this is that some states of the United States have either strong, moderate, weak or no law regarding the equal pay right protection between the wages of males and females (see Appendix 2). Therefore, the state of the firms headquarter is used to see if there is an effect on the total compensation of the CEO. Comparing to related literature about this topic, the state of the firm has not been used before and will be explored for the first time in this research.
To answer the research question of this paper, the Wharton Research Data Services, Compustat and ExecuComp databases have been used. The observations contain 88 American S&P500-Indexed firms, with relevant information on whether firms either have a male CEO or a female CEO and their total received compensation over the fiscal year 2016.
Based on previous literature, I expect that we can speak of a gender wage gap between male and female CEO’s. The results of this study, however, show that female CEO’s are being compensated with a higher amount than male CEO’s are earning in 2016. As the results show, being a female CEO is positively correlated with the total compensation. This outcome is, however, statistically insignificant. Even when insignificant control variables were removed, the coefficient of the gender variable remained insignificant. This implies that the gender wage gap in this sample is not present.
In the next chapter, a related literature review will be provided. In that chapter, the importance of the firm performance is discussed and possible explanations for the existence of the CEO gender compensation gap is given. In the third chapter, the research methodology is discussed which includes the research model, the data collection process, the explanations of the variables and at least the research method. In the fourth chapter, the results of the regression analysis are presented and will be discussed. In the fifth and last chapter, an overall conclusion will be given and thereafter a discussion and suggestions for further research.
Chapter 2: Related Literature
In section 2.1 we discuss evidence for and against the existence of the gender pay gap and for the remaining of the chapter we discuss possible explanations that have been proposed in the literature.
2.1 CEO compensation gap
US data from the 2012 Current Population Survey and ExecuComp show that even though women are a little more than 50% of white collar workers, they represent only 4.6% of executives (Flabbi et al., 2015). This finding can be partially explained by the fact that women earn less than men, ceteris paribus. When the General Motors male CEO, Daniel Akerson, resigned his position as a CEO in 2014, a woman called Mary Barra took over his position in the firm and became the new CEO of General Motors. Surprisingly, Barra earned $4.4 million, while Akerson was compensated $9 million. Even when Akerson stepped down to an advisory role, his compensation remained higher than Barra’s (Davis, 2015). This example shows that although researchers have found evidence for the existence of the gender wage gap, this example cannot be generalized to provide conclusive evidence.
Using a dataset on compensation and characteristics of CEO’s and other corporate officers, Newton and Simutin (2012, p.24) showed that the gender of the CEO, besides the age of the CEO, is an important determinant of this wage inequality. The explanatory variables: gender, age, years of tenure and the return on assets (ROA) for profitability reasons were included in their model to investigate the gender compensation gap. Bertrand and Hallock (2001, p.4) used the ExecuComp dataset, which contains information on compensation for the five top executive’s jobs of all firms who are part of the S&P MidCap-400, S&P-500 and the S&P SmallCap-600 for the years 1992 until 1997. The results of this analysis showed a gender pay gap of 45%, implying that women earn 45% less than men (Bertrand & Hallock, 2001, p. 17). A more recent study of Bell (2005, p. 4) has been using the same data as Bertrand and Hallock, only for the years 1992 until 2003. Thereby extending analysis to a period of greater participation of women executives generally and specifically in higher ranks (Bell, 2005, p. 3). Bell (2005, p.6-7) proposes that the gender gap in executive compensation has narrowed considerably through the late 1990s and the early 2000s periods, because of the greater executive participation rate of women. The results showed that women executives are paid between 8% and 25% less than male executives and that this effect is statistically significant
even when controlling for characteristics of the firm (market valuation, employment), executive (age, tenure and title) and industry known to influence compensation (Bell, 2005, p. 23).
On the contrary to the previous findings, in a research executed by Bowlin, Renner and Rives (2003, p. 751) the results indicated not enough statistical evidence for the executive gender pay gap to exist, using data of both male and female executives of the S&P-500 database over the year 1997. In another empirical research, Bugeja, Matolcsy, Spiropoulos (2012, p. 6) used data for firm in the US between 1998 and 2010. Their results also concluded that there exists no difference in compensation between male and female CEO’s (Bugeja, Matolcsy & Spiropoulos, 2012, p.16).
In a study done by Gayle, Golan and Miller (2012, p.833) the results concluded that women earn higher compensation than men, controlling for executive rank, background and experience, implicating an opposite gender pay gap. The reason for this, according to the authors, is largely due to women at all ranks and experience levels would exit a firm in higher rate than men. Female executives would gravitate to higher ranks and spend less time investing in human capital. This would explain the gender wage gap (Gayle, Golan & Miller, 2012, p. 866). This study consists of data on the 2,818 firms from the December 2006 version of the ExecuComp database and also background data of 16,300 executives by matching the 30,614 executives from the COMPUSTAT database for 1991 until 2006 with the records in the Marquis Who’s Who database (Gayle, Golan & Miller, 2012, p. 833).
2.2 Firm performance
In order to see whether the gender of the CEO affects the total compensation, it is most important to take into account an indicator of the overall performance of the firm. An important insight about an effective performance measure came from Jensen & Murphy (1991). In their paper, Jensen and Murphy (1991) argued that the ROA serving as a performance measure is relevant when determining the executive compensation. The paper incorporates data on thousands of CEO’s and provides information on salaries and bonuses for 2,505 CEO’s in 1,400 publicly held firms from 1974 through 1988 (Jensen & Murphy, 1991). In a theoretical study done by Hagel, Seely Brown and Davison (2010) it is argued that the best way to measure firm performance is to apply the ROA. The reason for this, according to the paper, is because the ROA explicitly takes into account the assets used to support
business activities (Hagel, Seely Brown & Davison, 2010). Besides that, Hagel, Seely Brown and Davison (2010) say that the ROA may foster a better view of the fundamentals of the business than other measures. As for my research, the return on assets will be used as a performance measure based on the findings of these previous studies.
2.3 Risk preferences
One of the most commonly used arguments explaining the gender wage gap are the differences in risk preferences between men and women. A CEO must ideally be able to take risks, failure to take risk is likely to force the firm to lag behind other competitors (SYN, 2011). Theory and empirical research are telling that women tend to be more risk-averse than men. This has been proven by the experimental results from the research of Powell and Ansic (1997). It presents the result of two computerized laboratory experiments designed to examine whether differences in risk preference and decision strategies are explained by the framing of tasks and level of task familiarity to subjects (Powell & Ansic, 1997). According to Powell and Ansic (1997), males and females also differ in adopting financial strategies. However, this difference has no significant impact on the ability to perform (Powell & Ansic, 1997). In another laboratory experiment using monetary incentives, Fehr-Duda, de Gennaro and Schubert (2006) concluded that women tend to be more risk-averse under specific circumstances. This could be due to women being less sensitive than men to probability changes in riskier investments and therefore also underestimate large probabilities of gains more strongly (Fehr-Duda, de Gennaro & Schubert, 2006).
In a paper of Dohmen and Falk (2011), the process of self-selection of a worker into different wage schemes is studied. The discussed wage schemes are either a fixed payment, independent of output, or a variable payment, dependent on output, consisting of three different treatments: a piece-rate, a two-person tournament or a revenue-sharing scheme. Variable payment schemes generally attract fewer women, an effect that is partly driven by an underlying gender difference in risk attitudes (Dohmen & Falk, 2011, p.583). The effect is strongest in the most competitive scheme, the tournament, where you either win the prize or lose everything relative to the performance of the opponent (Dohmen & Falk, 2011, p. 583). This indicates that women are risk-averse and also avoid competition. Following on the paper of Dohmen and Falk (2011), another paper of Niederle and Vesterlund (2007) studied whether women shy away from competition. Participants in a laboratory experiment solve a real task, first under a noncompetitive piece rate and then a competitive tournament incentive scheme
and although there are no gender differences in performance, men select the tournament twice as much as women when choosing their compensation scheme for the next performance (Niederle & Vesterlund, 2007, p.1067). Clustering the studies of Dohmen and Falk (2011) and Niederle and Vesterlund (2007), where women dislike competition and therefore tend to choose a fixed wage scheme over a tournament scheme, are interesting findings for the explanation of the gender wage gap because this would lead to lower average wages for women than for men (Dohmen & Falk, 2011, p.577). Because CEO’s should take risks wisely and be able to perform in a competitive environment, these factors could be a highly reasonable explanation for the difference in CEO compensation between genders.
2.4 Other explanations
Another explanation for the gender wage gap might be due to a difference in social preferences between genders. According to a theoretical study done by Croson and Gneezy (2004) there is evidence suggesting that women are more ‘other-regarding’ than men and therefore are more likely to choose professions which create benefits to other instead to themselves, which are also professions which are traditionally lower-paid (Croson & Gneezy, 2004, p.39). This explains why there are so few women incorporated as executives and why they could be compensated less than men.
When president John F. Kennedy implemented the Equal Pay Act in 1963, women earned on average 59 cents for every dollar a man was making (Cho & Kramer, 2013, p. 5). Because of this new legislation, to close the gender wage gap, the gap narrowed in 53 years to women earning 80 cents on average for every dollar a man is making (AAUW, 2017). To see whether legislation has an influence on closing the gender pay gap a comparison between a state with no equal pay law and a state with strong equal pay law should be made (see Appendix 2). A state with no equal pay law is Texas. The gender wage gap in Texas for 2016 showed that women on average earned 79 cents for every dollar a man was making (National Partnership for Women and Families, 2016). New York is a state with strong equal pay law and the gender wage gap for 2016 showed that in this state women earned on average 87 cents for every dollar a man was making (NPFWF, 2016). This indicates a difference of 10 cents. It is showing that the gender wage gap could also be explained by difference in legislation. Therefore, gender difference in CEO compensation could also be due to equal pay right protection provided by the state where the firm is located. To see an overview of the gender wage gap of more states in the U.S. see Appendix 3.
Economists Blau and Kahn (2007) have found that there existed an ‘unexplained’ gender pay gap in 2007 of 41%, which was due to unexplained factors and that these ‘unexplained’ factors could to some extent be due to gender discrimination in the labor market. In an earlier paper of Blau and Kahn (2000, p.96), they mention that women continue to confront gender discrimination in the labor market, although its extent seems to be decreasing. Even nowadays, women are still responsible for the child care in American households. Therefore, at least some of the gender pay gap is surely tied to the gender division of labor in the home, both directly through its effect on women’s labor force attachment and indirectly through its impact on the strength of statistical discrimination against women (Blau & Kahn, 2000, p.96-97).
Chapter 3: Research Methodology
This chapter will describe the research methodology used to analyze the data and answer the question whether we can speak of a gender wage gap between female and male CEO’s, or not, looking at S&P500-Indexed firms. First, the econometric model and statistical hypotheses will be provided, followed by an explanation about the functional form of the model. Second, the data collection process will be discussed. Third, the research variables, both dependent and independent, will be described. Finally, the last subsection will walk the reader through how the research will be executed.
3.1 Research Model
The econometric model that will be used throughout this research will be as follows:
TCi = 0 + 1Femalei + 2Agei + 3Age2i + 4Tenurei + 5Tenure2i + 6Statei + 7ROAi + i
In section 3.4 an explanation and arguments will be given why these particular variables are used and how they fit in the econometric model.
3.1.1 Statistical Hypotheses
Based on the findings of Newton & Simutin (2012), Bertrand & Hallock (2001) and Bell (2005), it is expected that the gender wage gap exists in our dataset when comparing the compensations of female- and male CEO’s. Therefore, the statistical hypothesis are as follows:
Null hypothesis: H0: 1 = 0 Alternative hypothesis: H1: 1 0
The null hypothesis implies that the coefficient of the female variable has no linear relationship with the dependent variable total compensation. The alternative hypothesis implies that the coefficient of the female variable does have a linear relationship with the dependent variable total compensation. If the coefficient is zero and the null hypothesis is rejected, then this does mean that the variables are independent, however not linearly dependent. Therefore, it is expected that the null hypothesis will be rejected against certain significance levels. This will be checked for a significance level of 1%, 5% and 10%.
3.1.2 Functional Form
When checking for the normality of the data, the data is not perfectly normally distributed. Because of this, the natural logarithm of the dependent variable total compensation is applied in a few papers (Bertrand & Hallock, 2001 and Bugeja, Matolcsy & Spiropoulos, 2012) to improve the normality of the data. However, when checking the normal distribution using the natural logarithm of total compensation, the data looks even more skewed and asymmetric than is in the first case. For this reason, the natural logarithm of total compensation will not be applied, because total compensation seems to approximate the normal distribution more closely (see figure 1).
Figure 1: Normal distribution of the dependent variable total compensation and the natural logarithm of total compensation
The tight green line represents the perfectly normal distribution, whereas the kinky green line represents the normal distribution of the data in reality.
3.2 Data collection
The data is obtained from Wharton Research Data Services (WRDS), Compustat and ExecuComp. ExecuComp offers the broadest view of the gender pay gap among top U.S. executives in publicly traded firms because the firms in this database constitute over 80% of the total market capitalization of U.S. public firms (Bell, 2005, p.5). The sample includes information on the name of the CEO’s, company names, total compensation of the CEO (including salary, bonus, other annual benefits, restricted stock grants, LTIP payouts, all other benefits and value of option grants), date CEO became CEO of the firm, date CEO left as CEO of the firm (if applicable), gender, age, state where the firm is headquartered, return on assets (ROA) of the firms. In total, 88 firms have been used which are all part of the S&P500-index, after removing outliers (see section 4.1.1). The data provides 19 companies with a
female CEO and 74 companies with a male CEO. For all the observations, the fiscal year 2016 is used because it is the most recent available data in ExecuComp. Besides that, 2016 is used because of simplifying reasons due to the fact that data during macroeconomic shocks or crises in previous years could have affected the data and might have provided ambiguous results. Therefore, previous data is excluded.
3.3 Research variables
3.3.1 Dependent variable
The dependent variable is the total compensation of a CEO in the fiscal year 2016. Total compensation includes: salary, bonus, other annual benefits, restricted stock grants, LTIP payouts, all other benefits and value option grants, as mentioned before in section 3.2. All these components are included because they are all part of the compensation structure of the CEO. Total compensation is used as the wage indicator instead of base salary alone, because all the components could be influenced by gender. Besides that, the S&P500 consists of firms of different industries. Therefore, the share in additional compensation differs across industries and because of that, base salary alone would not be a good indicator of the potential gender wage gap.
3.3.2 Independent variables
There will be six independent variables used in the econometric model. The model consists of one explanatory variable, CEO gender and of five control variables. Explanatory variables are necessary for explaining the model and reducing the variance of the error term. Control variables are also important to take into account because it is necessary to control for an omitted factor that determines the dependent variable (Stock and Watson, 2015, p.818).
Variable name Variable type Variable value
Total Compensation Continuous [0,]
Female Binary {0,1}
Age Continuous [0,67]
Tenure Continuous [0,17]
State Binary {0,1}
ROA Continuous [-0.0475, 0.154]
Since years of age and tenure are time indicators, it is assumed that age and tenure are continuous variables. The value of 67 and 17 is the maximum observed years of age and tenure in the data. -0.0475 is the minimum value of ROA observed in the data and 0.154 the maximum value.
Following the table with the description of the variables, explanations of why the variables were chosen is provided.
The first variable of the model is the most relevant one, namely the gender of the CEO, as it will provide an answer to my research question. This represents a dummy variable where the value of 1 implies that the CEO is a female and 0 if the CEO is a male. The second variable is used as a control variable and represents the age in years of the CEO. According to the findings of Bouvier (2010), age is a significant determinant for the compensation of the CEO. A positive relationship between age and compensation is expected. A third variable shows a quadratic term of age and is also added since the older you get, the less you might be able to work or perform as before when you were younger, affecting total compensation negatively (see figure 2 below). The fourth variable represents the years of tenure of the CEO and is also a control variable. Tenure is defined as years the CEO spent working as a CEO in the company. Tenure is calculated by deducting ‘the date left as CEO’ from ‘the date the CEO began in the firm’. Most CEO’s in the sample showed no end date, because they are still operating as CEO in the firm. Therefore, it is assumed that the end date for the sample CEO’s not showing an end date to be December 31, 2016, since this research only looks at the total compensation in the fiscal year 2016. In a study of Cordeiro and Veliyath (2003) it is stated that the relationship between years of tenure and total compensation is positive, indicating that the higher the years of tenure, the higher the level of compensation. However, this relationship can also be negative since tenure can have a negative effect on performance and, in the end also on compensation when the years of tenure increase and approach retirement. This negative effect indicates a quadratic relationship representing the next and fifth variable, the square of years of tenure. The sixth variable State is another dummy variable where the value of 1 implies that the firm of the CEO’s headquarter is located in an American State with moderate/strong equal pay right protection laws. A value of 0 implies a state with no/weak equal pay right protection. This variable is controlling to see whether the potential compensation gap is affected by these equal pay right legislations. The last control variable is ROA, the return on assets. As mentioned before in section 2.2, using a performance measure
is relevant when determining the compensation of an executive and that the best way to measure performance is to apply the ROA.
Figure 2: Scatterplots of the variables age and tenure against total compensation
The fitted line in the scatterplots show the negative relationship of age and tenure between total compensation.
3.4 Research method
A multiple linear regression analysis will be the method applied to address our research question. In order to estimate the coefficients of the independent variables and its necessary p-values, the Ordinary Least Squares method, or abbreviated OLS, will be performed, on the econometric model shown in section 3.1. However, before we can start estimating, OLS method relies on certain assumptions that we need to check before we can obtain reliable results. In case some assumptions are not fulfilled, the interpretation of the results can be slightly biased and inconsistent with reality and therefore (partially) untrustworthy. We will go through those assumptions in section 4.1. After verifying that our data satisfy the required assumptions, we can perform the regression analysis and obtain the results. Finally, we can start interpreting the results by testing whether the null hypothesis can be accepted or rejected using t-tests.
Chapter 4: Results
In section 4.1 we discuss the validity of the multiple linear regression assumptions and for the remaining of this chapter we show descriptive statistics, the results of this research and its discussion.
4.1 Validity of the multiple linear regression assumptions
There are six assumptions that needs to be fulfilled, which we can divide into two groups. The first group consists of assumptions we have to test before we start running the regression and the second group consists of assumptions we have to test after we have run the regression. We start with testing the assumptions of the first group:
4.1.1 No outliers
Outliers are exceptionally large or small values of the random variables (Stock & Watson, p.822, 2015). Removing outliers out of the data is important because not removing outliers will cause a skewed distribution of the data, where the OLS assumptions assume that normality of the data is required. The number of observations this research has started with was 105 observations. However, after removing outlier observations, the sample size reduced to 88, because the total compensation, age, tenure and ROA variables showed outliers which had to be removed (see appendix 4). Therefore, the first assumption is fulfilled.
4.1.2 Linearity
The second assumption is that the relationship between the dependent variable, total compensation and the independent variables is required to be linear. This can be checked using scatterplots in STATA. All the independent variables showed a linear relationship with the dependent variable total compensation (see figure 3). The first assumption is therefore fulfilled.
Figure 3: Scatterplots of the independent variables against total compensation
The upper scatterplot corresponds to the scatterplot of the ROA against total compensation, the scatterplot on the left corresponds to the scatterplot of age against total compensation and on the right, the scatterplot of tenure against total compensation is shown.
4.1.3 No multicollinearity
The third assumption of the multiple linear regression is that multicollinearity should be absent in the data. One way to check for multicollinearity is to create a correlation matrix of all the independent variables. Multicollinearity exists when two independent variables have a correlation above 0.80.
Table 2: Correlation matrix
The variable tenure squared is excluded from the x-axis since the correlation between tenure squared and tenure squared is one.
As is shown in table 2, none of the above independent variables have a correlation above 0.80. The variables age and tenure are positively correlated with each other and this is in line with the fact that the years of tenure might also increase when you get older. Despite the
hypothesis that female CEO’s get paid less than their male counterparts, the correlation between female and the dependent variable total compensation show a positive correlation of 0.1163, which means that female CEO’s would get paid more than male CEO’s in the sample. The assumption of no multicollinearity is therefore fulfilled.
For the remaining of section 4.1, we will test the last three assumptions which needs to be tested after running the regression:
4.1.4 Normality
Normality of the residuals is the second assumption that needs to be examined. A residual is the difference between the observed value of the dependent variable and the value of the dependent variable predicted by the econometric model (Stock & Watson, p.822, 2015). Residuals can only be estimated after running the regression. Normality of the residuals will be checked by using a Jarque-Bera test with a significance level of 5%. The null hypothesis of this test states that the residuals are normally distributed. The resulting p-value of this test is 0.195, which is greater than 0.10, 0.05 and even 0.01, indicating that the null hypothesis cannot be rejected for these significance levels. Therefore, the residuals are normally distributed and the fourth assumption is fulfilled.
Figure 4: Normal distribution of the residuals
A histogram of the residuals of the sample with the kinky green line representing to what extent the residuals are normally distributed.
4.1.5 Homoscedasticity
The fifth assumption of the multiple linear regression is that the variances of the error terms are constant across all the values of the independent variables, which is called homoscedasticity. To check for homoscedasticity, a Breusch-Pagan test will be performed with a significance level of 5%. The null hypothesis of this test states that we can speak of homoscedasticity. A p-value lower than 0.05 will reject the null hypothesis and therefore
implies heteroscedasticity. The obtained p-value is 0.5009, which is greater than 0.10, 0.05 and 0.01, meaning that homoscedasticity cannot be rejected at every significance level. The sample variances of the error term are constant across all values of the independent variables and are therefore homoscedastic (see figure 5 below), fulfilling the fifth assumption.
Figure 5: Homoscedasticity
A scatterplot of the residuals against the fitted values of the sample with each dot representing the variance of the error term of the value of the independent variables.
4.1.6 No autocorrelation
The final assumption is that there exists no autocorrelation. Autocorrelation means whether the residuals are correlated with previous periods. To test for autocorrelation, a Durbin-Watson test is performed. The null hypothesis of this test states that there is exists no autocorrelation whereas the alternative hypothesis states that there is positive autocorrelation. A value near two of the d-statistic indicates that we cannot reject the null hypothesis. In this case, the d-statistic is 1.936853, which means that the last assumption of no autocorrelation is fulfilled.
4.2 Descriptive Statistics
Table 3: Descriptive statistics
The number of observations in this research are 88 over the fiscal year 2016. Both dependent and independent variables are included in this table representing their mean sample value, the standard deviation and their maximum and minimum sample value.
Table 3 provides summarizing statistics for the independent variables and the dependent variable. The percentage of 21.6 percent indicates that the sample consists of 19 females and 59 males. A CEO originated from the sample is on average 57 years old, with the youngest CEO in the sample being 45 years old and the oldest being 67 years old. The average years of tenure approximately is 5.44 years, with some CEO’s in the sample showing no tenure at all and 17 years of tenure being the highest level of tenure observed. The variable state is a dummy variable where the value of 1 indicates that the firm of the CEO is headquartered in a state of the United States with strong or moderate equal pay right law protection. The mean of this variable shows a percentage of 28.4 percent, implying that 25 of the 88 firms headquarter is located in a state with strong or moderate equal pay right protection. The remaining observations are dealing with no or weak equal pay right protection. The average return on assets in the sample is 4.99 percent, with a negative minimum value of 4.75 percent and a positive maximum value of 15.44 percent. Finally, the dependent variable, total compensation of the CEO, shows on average a value of $11,771,440. A CEO earning $25,168,600 obtained the highest amount of total compensation in 2016 and a CEO earning $283,020 obtained the lowest amount of total compensation in 2016.
4.3 Regression and discussion
The following table, table number 4, provides the outcomes and the results of the regression analysis.
Table 4: Results of the multiple linear regression analysis
The in red underlined p-values are the p-values greater than 0.10 (significance level of 10%) for which the variables are statistically insignificant. The in blue underlined p-value is the p-value smaller than 0.10 and 0.05 but greater than 0.01 (significance level of 1%). The p-values who are not underlined are the p-values that are smaller than 0.01.
Table 4 shows the outcomes of the multiple linear regression analysis. The most important independent variable for my research question is the female variable, to see whether total compensation depends on being a male or a female. The positive sign in front of the coefficient of 1533.65 implies that in this sample, female CEO’s are compensated a higher amount of money than male CEO’s in the year 2016 with approximately $1,533.65. This result is the opposite of what was expected to be the outcome as stated in the hypothesis, where male CEO’s were expected to be compensated more than the female CEO’s in the sample, resulting in a negative sign in front of the female variable coefficient. However, the p-value of 0.193 indicates that this result is statistically insignificant when checking for a significance level of 0.10, meaning that we cannot reject the null hypothesis (see section 3.1.1) that the coefficient of the female variable is equal to zero, implying that in this sample, we cannot speak of a gender wage gap between male and female CEO’s. The age of the CEO
is also statistically insignificant when checking for a significance level of both 0.01, 0.05 and 0.10 meaning that the age of the CEO also has no significant effect on the total compensation. However, an additional year of age seems to increase the total compensation of the CEO with $1,725.94 at the same time, the effect of age diminishes with respect to an additional year with $14.83. This effect is due to the fact that as you get older, your ability to work will slow down and negatively affect your earnings, but in this case, the age squared variable is also statistically insignificant for every significance level. The intuition behind age being an insignificant determinant of the total compensation of the CEO is that CEO’s already get compensated a high amount compared to other working positions. Becoming one year older and besides that working in the highest possible position will most likely have little effect on your income. An additional year of tenure seems to increase total compensation with $1,199.79, but this effect also diminishes with respect to an additional year of $61.09. This increasing effect caused by tenure is statistically significant when checking for a significance level of 0.10, 0.05 and 0.01. This implies that we can reasonably assume that the years of tenure do increase the total compensation of a CEO by $1,199.79 per extra year. The diminishing effect caused by the years of tenure squared is statistically significant when checking for the significance level of 0.10, 0.05 and 0.01, which means that we can assume that an extra year of tenure squared decreases total compensation with $61.09 per year. The coefficient of the state variable shows that a CEO’s whose firm is headquartered in a state with strong or moderate equal pay law is compensated $2,447.58 higher than is the case in a state with weak or no equal pay law. This effect is significant when checking for the significance levels of 0.10 and 0.05. Because of this, we can assume in this case that the state where the firm is headquartered, regarding equal pay law, is significant enough to have an effect on the total compensation. The independent variable, return on assets, is highly insignificant because its p-value of 0.593 is much larger than even 0.10. This means that in this sample, the return on assets has no impact on the total compensation of the CEO. This could be due to the fact that the return on assets is also affected by factors outside the control of the CEO, causing the return on assets to have an insignificant effect on compensation. However, this could also mean that return on assets as a performance measure might have been the wrong choice when controlling for firm performance in this sense. The R-squared is 0.2213, which means that 22.13 percent of the variation in the model was explained by the model, the rest remaining unexplained. This percentage could have been higher when more control variables would have been included in the model.
Removing the insignificant control variables out of the regression analysis, will reduce the standard errors of the remaining variables. In this way, the female variable might become significant when the insignificant control variables are removed.
Table 5: Regression analysis without insignificant variables age, age squared and ROA.
The in red underlined p-value is the p-value greater than all significance levels. The p-values who are not underlined are the p-values which are significant for the significance levels of ten and five percent.
In table 5, the coefficient in front of the female variable still shows a p-value larger than 0.10, indicating that also in this case, the female variable is not significant for every significance level. Even after removing the insignificant variables and thereby reducing the standard errors, the results show that we cannot speak of a gender wage gap between male and female CEO’s looking at S&P500-indexed firms in 2016.
Chapter 5: Conclusion, discussion & suggestions for further research
In section 5.1, a summary of the findings will be given, followed by the conclusion of this research. For the remaining of this chapter, a discussion of this research and its limitations will be provided and thereafter suggestions for further research will be discussed.
5.1 Conclusion
In chapter 4, regression analysis provided us with the means to give an answer to the question whether we can speak of a gender wage gap when looking between male and female CEO’s at S&P500-indexed firms in the year 2016. The positive coefficient of the female variable contradicts the hypothesis that male CEO’s are compensated a higher amount than female CEO’s. As mentioned before in section 4.3, the insignificance of the female coefficient indicates that we cannot speak of a gender wage gap in 2016, because being female does not appear to be a significant determinant of a CEO’s compensation. This could mean that discrimination is less likely to happen nowadays than before, since equal pay law do have an effect on the total compensation of the CEO and as the paper of Bell (2005) showed that the gender wage gap was more closing than it was at the time when Bertrand and Hallock (2001) found a gap of 45% and that this could be due to the higher participation rate of female executives, which is still rising today. Since only four female CEO’s were found in the S&P500-index in 2001 and 2005, and 23 female CEO’s in 2016 (WRDS). To be absolutely able to conclude that being a female does not play a role in CEO compensation, more female CEO’s must be examined and there must be available sufficient information about the CEO’s and their environment, which is both not the case.
5.2 Discussion
There are some potential limitations regarding this study that needs to be discussed. First of all, the number of female CEO’s in the sample is very limited compared to the number of male CEO’s. As a consequence, the number of observations, compared to most studies, is very small and therefore we lack the statistical power to obtain significant results. Having access to more female CEO’s in a database would make the results more accurate and reliable, which brings me to my next limitation, namely the two unsatisfied OLS assumptions, normality of the residuals and homoscedasticity of the variances of the residuals. More observations must be done in order to improve the normal distribution of the data. The consequences of heteroscedasticity in this research was fixed by using the robust option for calculating the standard errors. Another limitation could be that this research only focuses on
a few variables that could have an effect on the total compensation, with gender as the predicted main factor, whereas for example, years of education, also could have been a significant factor. Studies of Blau & Kahn (2007) and Gayle, Golan & Miller (2012) control for years of education. The reason why this variable is excluded in this research, is because working as a CEO is the highest possible position one could achieve and therefore I assume that every CEO has attained to some extent similar years of education and for this reason, years of education is neglected. Besides that, in a study done by AAUW (2017), it is argued that education is not an effective explanation for pay gap. While more education is a useful tool for increasing earnings, it is not effective against the gender pay gap and at every level of academic achievement, women’s median earnings are less than men’s earnings and in some cases, the gender pay gap is larger at higher levels of education (AAUW, 2017). Quality of education is, in my opinion, more relevant, because going to school for the same amount of years, will say nothing when one CEO has attained high quality education while the other has not. However, this variable was not available. Finally, bonus and fixed salary could have been used as two separate dependent variables. Using bonus as a dependent variable could have checked whether it is to some extent true that female CEO’s gain high fixed salary and low bonuses compared to a male CEO’s compensation, assuming the fact that women are more risk averse than men (see section 2.3).
5.3 Suggestions for further research
One direction for further research could be taking an additional executive level, lower than the CEO’s position, into consideration and make a comparison between the two levels and see in which executive level the pay gap is more of a problem. Another interesting direction for further research is to check for every compensation component to what extent they are affected by the gender of a CEO. The researcher could then recommend more integrating remuneration policies for the CEO’s compensation to The Remuneration Committee. In my opinion, it is very interesting to investigate the gender wage gap in different countries, to see whether ethnicity might have a significant effect on the gender wage gap.
References
American Association of University Women. (2017). Simple Truth About the Pay Gap
Spring 2017 Edition. Retrieved from: h ttps://www.aauw.org/research/the-simple-truth-about-the-gender-pay-gap/
Blau, F.D., & Kahn, L.M. (2000). Gender Differences in Pay. Journal of Economics
Perspectives, 14(4), 75-99.
Blau, F.D., & Kahn, L.M. (2007). The Gender Pay Gap: Have Women Gone as Far as They Can? Academy of Management Perspectives, 21(1), 843-860.
Bell, L.A. (2005). Women-led Firms and the Gender Gap in Top Executive Jobs. IZA
Discussion Paper, no. 1689. Institute for the study of labor, Bonn.
Bertrand, M., & Hallock, K.F. (2001). The Gender Gap in Top Corporate Jobs. Industrial and
Labor Relations Review, 55(1), 3-21.
Bouvier, A. (2010). The Effect of Age upon CEO Compensation: A Cross-Industry Study. Retrieved from:
http://scholarship.claremont.edu/cgi/viewcontent.cgi?article=1016&context=cmc_thes es
Bowlin, F., Renner, J., & Rives, M. (2003). A DEA Study of Gender Equity in Executive Compensation. The Journal of The Operational Research Society, 54(7), 751-757.
Bugeja, M., Matolcsy, Z., & Spiropoulos, H. (2012). Is there a Gender Gap in CEO Compensation? Journal of Corporate Finance, 18(4), 849-859.
Cho, R., & Kramer, A. (2013). Everything You Need to Know about the Equal Pay Act.
International Center for Research on Women. Retrieved from:
https://www.icrw.org/wp-content/uploads/2016/11/Everything-You-Need-to-Know- about-the-Equal-Pay-Act.pdf
Determinants of CEO Compensation. American Business Review, 21(1), 56-66.
Croson, R., & Gneezy, U. (2004). Gender Differences in Preferences. Journal of Economic
Literature, 47(2), 448-474.
Davis, O. (2015). Why Women Executives Get Paid Less Than Male Counterparts – New Fed Study Explains. International Business Times. Retrieved from:
http://www.ibtimes.com/why-women-executives-get-paid-less-male-counterparts- new-fed-study-explains-1858976
Dohmen, T., & Falk, A. (2011). Performance Pay and Multidimensional Sorting:
Productivity, Preferences and Gender. American Economic Review, 101(2), 556-590.
Fehr-Duda, H., De Gennaro, M., & Schubert, R. (2006). Gender, Financial Risk and Probability Weights. Theory and Decision, 60(2), 283-313.
Flabbi, L., Mario, M., Moro, A., & Schivardi, F. (2015). Female Executives and Gender Wage Gaps. Retrieved from: http://voxeu.org/article/female-executives-and-gender-wage-gaps
Gayle, GL., Golan, L., & Miller, R.A. (2012). Gender Differences in Executive Compensation and Job Mobility. Journal of Labor Economics, 30(4), 829-872.
GlassCeiling. (2015). Exploring the Gender Pay Gap. Retrieved from: http://www.glassceiling.com/exploring-the-gender-pay-gap/
Hagel, J., Seely Brown, J., & Davinson, L. (2010). The Best Way to Measure Company Performance. Harvard Business Review. Retrieved from: https://hbr.org/2010/03/the-best-way-to-measure-compan
Jensen, M., & Murphy, K. (1991). CEO Incentives: It’s not how much you pay but how.
Harvard Business Review, 68(3), 138-153.
https://hbr.org/2009/05/what-only-the-ceo-can-do
National Partnership for Women & Families. (2016) New York Women and the Wage Gap. Retrieved from: http://www.nationalpartnership.org/research-library/workplace- fairness/fair-pay/4-2016-ny-wage-gap.pdf
National Partnership for Women & Families. (2016). Texas Women and the Wage Gap. Retrieved from: http://www.nationalpartnership.org/research-library/workplace- fairness/fair-pay/4-2016-tx-wage-gap.pdf
Newton, D., & Simutin, M. (2014). Of Age, Sex and Money: Insights from Corporate Officer Compensation on the Wage Inequality between Genders. Management Science,
61(10), 2355-2375.
Niederle, M., & Vesterlund, L. (2007). Do Women Shy Away From Competition? Do Men Compete Too Much? Quarterly Journal of Economics, 122(3), 1067-1101.
OECD. (2017). Gender wage gap (indicator). doi: 10.1787/7cee77aa-en (Accessed on 28 November 2017).
Powell, M., & Ansic, D. (1997). Gender Differences in Risk Behaviour in Financial Decision- Making: An Experimental Analysis. Journal of Economic Psychology, 18(6), 605-628.
Skills You Need. (2011). Want to be a CEO? Retrieved from: https://www.skillsyouneed.com/rhubarb/ceo-skills.html
Stock, J.H., & Watson, M.W. (2015) Introduction to Econometrics. Essex, United Kingdom: Pearson Education Limited.
The Economist Group Limited. (2017). The Gender Pay Gap. The Economist. Retrieved from: https://www.economist.com/news/international/21729993-women-still-earn-lot-less-men-despite-decades-equal-pay-laws-why-gender
Appendices
Appendix 1: Uncontrolled Lifetime Earnings by Gender
Source: GlassCeiling. (2015). Exploring the Gender Pay Gap. Retrieved from:
Appendix 2: States with Equal Pay Protections (Updated 06-07-2016)
Delaware, Minnesota and Oregon are assumed to be part of the states with strong or some form of major equal pay protections.
Source: Connel et al. (2016). Updated Maps: States with Equal Pay Protections and Pending Equal Pay Legislation. Orrick. Retrieved from:
Appendix 3: Comparing the gender wage gap among U.S. states1
States with no/weak equal pay law States with major/strong equal pay law
1] The above states are the ones available in the used data sample.
2] All numbers under the column ‘Wage Gap’ indicate how much cents a woman makes compared to one dollar a man makes in that particular state in 2016.
Sources: All the numbers displayed above are retrieved from the factsheets, for every state, from National Partnership for Women & Families. The factsheets are originated in April 2016.
3] National Partnership for Women & Families. (2016). New York Women and the Wage Gap. Retrieved from:
http://www.nationalpartnership.org/research-library/workplace-fairness/fair-pay/4-2016-ny-wage-gap.pdf
State Abbreviation Wage Gap
Alabama AL 0.732 Arizona AZ 0.84 Colorado CO 0.82 Connecticut CT 0.83 Florida FL 0.85 Georgia GA 0.82 Indiana IN 0.75 Louisiana LA 0.65 Michigan MI 0.75 Missouri MO 0.77 North Carolina NC 0.85 Nebraska NE 0.79 New Jersey NJ 0.80 Ohio OH 0.78 Pennsylvania PA 0.79 Texas TX 0.79 Virginia VA 0.80 Washington WA 0.77 Wisconsin WI 0.79
State Abbreviation Wage Gap
California CA 0.84
Illinois IL 0.79
Maryland MD 0.85
Minnesota MN 0.81
Appendix 4: Outliers
Boxplot of Total Compensation, where the dots in every Boxplot of Age boxplot represent the outliers.
Appendix 5: The performed test outcomes of 4.1.4, 4.1.5, 4.1.6
Jarque-Bera test for normality of the residuals (section 4.1.4)
Breush-Pagan/Cook-Weisberg test for homo- and heteroscedasticity (section 4.1.5)
