1
The relation between board gender diversity and corporate social
performance: insights from the European area
A.R.C.G. Bijstra
1Double Degree International Financial Management (IFM)
University of Groningen & University of Uppsala
Supervisors: Prof. Dr. C.L.M. Hermes and Dr. H. Vrolijk
16 June 2015
ABSTRACT
This paper examines the relation between board gender diversity and corporate social performance within 600 firms in 18 European countries over the period 2007 to 2014, while addressing differences in national culture. Since ordinary least squares estimates are downward biased, it is possible that reverse causality is a serious problem that should be circumvented. The two-stage least squares results demonstrate a positive relation between board gender diversity and corporate social performance in the post-crisis period 2011 to 2014. Furthermore, Hofstede's Masculinity/Femininity dimension variable positively moderates the relation between board gender diversity and corporate social performance, while the Individualism/Collectivism dimension variable does not moderate this relation. Several robustness tests substantiate the two-stage least squares results.
Keywords: Board of Directors, corporate social performance, gender diversity, national
culture, social role theory
JEL classification: G30, G34, J16, M14
1 Address: Wittgensteinlaan 342, 1062 KJ, Amsterdam (The Netherlands). E-mail addresses:
2
1. Introduction
Big corporate scandals like Enron, Parmalat, and WorldCom and the failures of financial institutions like the Lehman Brothers during the financial crisis have lowered confidence in the corporate world. Moreover, these have resumed the conversation on
corporate social performance (CSP)2 and have raised the question whether CSP outcomes would be different if Boards of Directors (BoDs) had other compositions (Terjesen et al., 2009). The narrow-minded focus of firms on financial outcomes and the lack of consideration of CSP outcomes in corporate decision-making are regarded as the most important reasons explaining the recent financial crisis (Souto, 2009). As the decisions of BoDs (with respect to CSP) affect the welfare of other stakeholders than merely the shareholders, it seems hard to disregard a view arguing that the composition of BoDs should be more aligned with diversity of society in general to guarantee that decisions are taken with a wider (i.e., stakeholder orientation) view (Rose, 2007). From the perspective of the government this wider view makes sense, as paying attention to merely financial performance (FP) is clearly suboptimal for society as a whole. Therefore, a lot of governments have focused on raising board gender
diversity (BGD) to increase CSP and to prevent new corporate scandals and failures from
occurring. To reach this goal, they have already implemented or are in the creation phase of introducing legislation. For example, Norway enacted a gender quota law in 2003 requiring listed firms to have 40 percent female BoD members by 2008 (Adams and Funk, 2012). In Germany 30 percent of the BoD members of listed companies have to be women by 2016 (Daling, 2014). But why would a firm engage in (costly) activities that affect CSP outcomes? Aguilera et al. (2007) and Bansal and Roth (2000) argue that more and more firms are moving from their narrow-minded focus on financial outcomes towards a focus on financial, social, and environmental outcomes. The rationale is that the latter two could be a driver for innovation and competitive advantage, and hence ultimately FP. Consequently, the engagement in CSP-related activities fits with the raison d’être of the firm: the maximization of shareholder value (e.g., Fama and Jensen, 1983; Jensen and Meckling, 1976; Lazonick and O’Sullivan, 2000). While previous academic works on the relation between BGD and firm outcomes have primarily focused on FP, i.e., maximization of shareholder value (e.g., Carter et al., 2003; Erhardt et al., 2003; Rose, 2007), CSP has not received much attention. However, in the context of the increasing number of females in BoDs due to the worldwide trend to
3
enact BGD quotas and the increasing focus of firms on CSP outcomes, it is important to inquire the impact of BGD on the firms’ CSP.
Wood (1991) defines CSP as ‘a business organization’s configuration of principles of
social responsibility, processes of social responsiveness, and policies, programs and observable outcomes as they relate to the firm’s societal relationships’. A firm’s BoD that
gives high priority to CSP aims to improve social and environmental outcomes by dealing with issues like family leave, employee training and development, customer’s health and safety, resource reduction, and emission reduction (Kakabadse, 2007). This implies that these outcomes have potential benefits for other stakeholders (e.g., its employees, its customers, and society at large) than just the shareholders of the firm. Personal values of the BoD members are important in this respect, as they are a predictor of directors’ behaviors (Sagiv et al., 2011). A personal value is an abstract and desirable goal that a person strives to achieve and that occupies a crucial place within the social psychology of that person (Rokeach, 1973). Personal values that specifically underline concern for the welfare of other stakeholders than just the shareholders are crucial with regard to CSP. Boulouta (2013) and Hafsi and Turgut (2013) find that fundamental differences between values of male and female BoD members explain the positive relation between BGD and CSP within their samples of S&P500 firms. Hence, gender differences with respect to values of BoD members could help to predict whether and how CSP will change as the composition of the BoD with respect to gender changes.
Two questions are still unanswered though. First, although fundamental differences between the values of men and women exist in corporate America, the question is whether these differences exist in every firm worldwide. Countries like The Netherlands, Sweden, and Norway have national cultures that show a less clear distinction of values between men and women. Since values are the most fundamental features of culture and shape the behaviors of all individuals, cultural values also shape the behaviors of BoD members (Hofstede, 2001). Secondly, Elstad and Ladegard (2012) argue that the real influence of female BoD members is possibly restrained as they are part of the minority in the BoD. This restriction is more or less present depending on the cultural values prevailing in a country (Darwish and Huber, 2003; Hofstede, 2001). Hence, the second question is to what extent cultural values could alleviate or aggravate this restriction. Hofstede’s (2001) Masculinity/Femininity (MAS) and
Individualism/Collectivism (IDV) dimensions of national culture could answer both questions.
4
2007 to 2014. I use a two-stage least squares (2SLS) estimation method to tackle the potential problem of reverse causality. The instrument that I employ is the female full-time economic labor force participation (FFELFP), following Adams and Kirchmaier (2013) and Compton et al. (2014). In the second part I investigate whether, and if so to what extent, nation-level cultural institutions moderate the relation between BGD and CSP in the culturally diverse European area, as the 18 home countries where a firm could reside have divergent national cultures with respect to the MAS and IDV dimensions (Hofstede, 2001). Hence, my research adds to the existing literature regarding the relations between BGD and CSP (Boulouta, 2013; Hafsi and Turgut, 2013) as well as nation-level cultural institutions and CSP (Ioannou and Serafeim, 2012), by answering the question: does a more gender diverse board lead to higher CSP or is it contingent on national culture?
The remainder of the paper is structured as follows. Section 2 provides a literature review. Moreover, it discusses the inquiry’s hypotheses. Section 3 specifies the empirical analysis. Section 4 shows the results. Section 5 discusses and concludes the paper.
2. Literature review and hypotheses
In this literature review I provide theory that can be employed to investigate the relation between BGD and CSP. First, I touch upon the agency theory and the resource dependence theory to show the responsibilities of the BoD regarding the firm and hence its activities resulting in CSP outcomes. Then, I provide the theoretical lens which explains how gender differences with respect to values affect behaviors of men and women, namely the social role theory. Next, I discuss which feminine values that are important with regard to CSP outcomes persist to the BoD. Immediately thereafter I provide my initial hypothesis. Subsequently, I discuss the moderating effect of MAS on the relation between BGD and CSP, followed by my second hypothesis. Finally, I describe the moderating effect of IDV on the relation between BGD and CSP, followed by my third hypothesis.
2.1 The importance of the BoD
The agency theory points at the relation where in a contract ‘one or more persons (the
principal(s)) engage another person (the agent) to perform some service on their behalf which involves delegating some decision making authority to the agent’ (Jensen and Meckling,
5
(principals) of a company are not always aligned. The reason for this is the separation of ownership and control within organizations (Fama and Jensen, 1983; Jensen and Meckling, 1976). A company’s BoD can be seen as an internal control mechanism with the task to keenly control and monitor the behaviors of top managers (agents) to guarantee that they act in the shareholders’ (principals) interests (Eisenhardt, 1989). BoDs control and monitor the behaviors of top managers via a number of specific activities. Among other things, these activities include monitoring the firm’s financial reporting, evaluating and rewarding CEO and top managers, and evaluating strategic plans with regard to mergers and acquisitions (Hillman and Dalziel, 2003).
The resource dependence theory stresses that firms depend on scarce resources owned by external actors and organizations. The BoD has the ability to connect the firm with these external actors and organizations and to bring critical resources to the organization (Pfeffer and Salancik, 1978). Among other activities, the provision of resources function of the BoD comprises providing legitimacy, helping in the formulation of strategy, administering advice and counsel, connecting the firm to crucial stakeholders, and facilitating access to capital (Hillman and Dalziel, 2003). Personal networks of BoD members are crucial for these activities. To exemplify, if a firm has a BoD member who is also BoD member of a bank, it has a higher chance to be financed by that bank.
Thus, both the agency and the resource dependence perspectives argue that the BoD influences the firm’s outcomes via several activities, which implies that the BoD also plays a key role in influencing the firm’s CSP outcomes (Huse et al., 2009; Rosener, 1990).
2.2 Social role theory
6
investigate gender stereotypes in different populations and hence cultures all over the world. They link women to values like fellow feeling (empathy), caring, great concern for people around them, and being keen on substantiating values in relationships which are of great significance to the community. They find that women exhibit more socially facilitative behaviors than men, which could mean that women more successfully address social and environmental (i.e., CSP) issues like family leave, employee training and development, customer’s health and safety, resource reduction, and emission reduction than men. However, these findings do not imply that women in BoDs have the same values, being predictors of behaviors (Sagiv et al., 2011), as women in the general population. For that reason, it is necessary to discuss whether, and if so to what extent, these values persist to the boardroom.
2.3 BoDs, values, and CSP
The research by Adams and Funk (2012) sheds light on the question how core values of male and female directors differ from gender differences in the general population. In line with the findings of Eagly and Karau (1991) and Fox et al. (1985) in the general population, they show that female BoD members are more universally concerned and benevolent than their male counterparts. The definitions of universalism and benevolence show that both values are important with respect to CSP-related issues and its outcomes. While universalism is defined as ‘Understanding, appreciation, tolerance, and protection for the welfare of all
people and for nature’, benevolence is defined as ‘Preservation and enhancement of the welfare of people with whom one is in frequent personal contact’ (Schwartz and Rubel, 2005).
To conclude, male and female directors differ significantly with respect to the core values of universalism and benevolence, which are important determinants of socially facilitative behaviors. These behaviors are in turn important to successfully influence CSP-related issues and the outcomes thereof. Therefore, I argue that a higher degree of BGD positively affects CSP outcomes. I translate this reasoning into my initial hypothesis:
Hypothesis 1: BGD has a positive effect on the CSP of an organization3.
3 The maximum BGD is reached at a ratio of 0.5, as this is literally the most gender diverse ratio. Yet, in this
7
2.4 The moderating effect of MAS
Taking the aforementioned literature about social role theory into consideration, one could argue whether this theory is applicable to every single country in the world. The social role theory implies that the connection between BGD and CSP depends on the actual existence of fundamental differences between values of male and female directors. For that reason, in countries where stereotypical beliefs with respect to gender are not so strong, this theory might not be useful (Boulouta, 2013). According to (Hofstede, 2001), one can make a distinction between masculine and feminine societies. Whereas a masculine society ‘stands
for a society in which social gender roles are clearly distinct: men are supposed to be assertive, tough, and focused on material success; women are supposed to be more modest, tender, and concerned with the quality of life’ a feminine society ‘stands for a society in which social gender roles overlap: Both men and women are supposed to be modest, tender, and concerned with the quality of life’. In a feminine society, relationships, people, and
environment protection are high priorities (Hofstede, 2001). A good example of a feminine society, according to Hofstede (2001), is Sweden. This country has a score of 5 on the MAS dimension variable, whereas e.g., Austria has 79. Therefore, in Sweden both men and women are supposed to care about relationships, people, and environment protection, which implies two things. On the one hand, a low MAS score could represent a higher tendency to seriously address CSP-related issues, irrespective of how a BoD is set up regarding gender. On the other hand, one could also argue that in Sweden it might not matter for the CSP of the organization how the BoD is composed with respect to gender, as men and women behave rather similarly in contrast to Austria. As I want to examine the moderating effect of MAS on the relation between BGD and CSP, I regard the second implication to be important for my inquiry, signifying that the higher the score on Hofstede’s MAS dimension variable, the larger the impact of BGD on CSP. This reasoning brings forth my second hypothesis:
Hypothesis 2: The positive effect of BGD on CSP is stronger as the firm’s home country scores higher on Hofstede’s MAS dimension variable.
2.5 The moderating effect of IDV
8
actually have the opportunity to influence BoD decisions. As females remain a demographic minority regarding gender in BoDs all over the world, some scholars argue that females are still seen as an out-group by the demographic majority regarding gender in BoDs. This majority consists of the male members who are also called the in-group (Singh and Vinnicombe, 2004; Tsui et al., 1992). In the context of a BoD, there is proof that men of the in-group have a propensity to evaluate the behaviors of other men in the in-group more positively compared to similar behaviors in the out-group, which is the minority being women (Singh and Vinnicombe, 2004). Furthermore, women that are part of a minority in a BoD possibly have difficulties with regard to full participation in social interaction with other BoD members, since they identify themselves as an out-group (Huse and Solberg, 2006). As a result, it is reasonable to assume that women in BoDs face substantial challenges with respect to individual participation and influence on decision-making on the BoD. Thus, it is questionable whether women are really key influencers or just tokens in their BoDs (Elstad and Ladegard, 2012).
Also on this question Hofstede’s (2001) theory about national culture could shed light, more specifically his theory about the IDV dimension. As stated by Hofstede (2001), an individualistic society is characterized by high importance on individual initiative and achievement, same rights for everybody, and hiring and promotion built on skills. On the contrary, a collectivist society is characterized by belonging (or not) to in-groups, rights depending on the group where someone belongs to, and hiring and promotion based on in-group status. Good examples of collectivist societies, according to Hofstede (2001), are Portugal and Greece. These countries have low scores of 27 and 35 on the IDV dimension variable respectively, whereas e.g., The Netherlands has a relatively high score of 80. While relationships with colleagues (or BoD members) are cooperative for in-group members in Portugal and Greece, they are unfriendly or even hostile to members of the out-group. Personal relationships with in-group members are even more important than task and company (Hofstede, 2001). Darwish and Huber (2003) argue the following about the in-group in a collectivist society: ‘Their problem with the members of the (individualist) out-group
would be that they experience and categorize them as individualists—highly distant from their own experiences as members of a collective group and highly distinct from each other’.
9
to a similar minority of women in The Netherlands, since their individual initiative and achievement, rights, and skills are less important. Far more important are their belonging to the in-group (or not) and the rights and status connected to this in-group. Hence, a minority of women in a BoD in Portugal and Greece, which is not part of the in-group (Singh and Vinnicombe, 2004; Tsui et al., 1992), is probably more restrained to take part in CSP decision-making in the BoD. Thus, I argue that the higher the score on Hofstede’s (2001) IDV dimension variable, the larger the effect of BGD on CSP. This effect of Hofstede’s (2001) IDV dimension variable leads to my third hypothesis:
Hypothesis 3: The positive effect of BGD on CSP is stronger as the firm’s home country scores higher on Hofstede’s IDV dimension variable.
3. Empirical analysis
3.1 Sample and data
For my research I employ a dataset of companies which are included in the STOXX® Europe 600 Index as of 20 March 2015, covering the years 2007 to 2014. The STOXX® Europe 600 Index encompasses 600 constituents, which are publicly listed European companies. The index is representative of large, mid, and small capitalization firms across 18 countries of the European area, namely: Austria, Belgium, Czech Republic, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and the United Kingdom. As a result, this index offers the opportunity to compare several European countries with divergent scores on the MAS and IDV dimension of Hofstede. Lastly, the index represents approximately 90 percent of Europe’s total market capitalization. Thus, regarding both the representation of the number of countries as well as the percentage of the total European market capitalization, I consider the STOXX® Europe 600 Index to be a representative index for entire Europe.
Regarding the collection of the data, firm level data originate from Worldscope and
Thomson Reuters’ ASSET4 (ESG). I collect data on board characteristics, i.e., BGD and board
independence also via Thomson Reuters’ ASSET4 (ESG). As to country level data, which are the dimension scores of Hofstede and an instrument, I use the work of Hofstede (2001) and
Euromonitor respectively. For an overview of the definitions and data sources of the variables
10
3.2 CSP index (dependent variable)
In previous academic articles the construction of an accurate measure of CSP has been quite thought-provoking, since (1) the theoretical construct of CSP is multidimensional and (2) measurements of a single feature of CSP (e.g., CO2 emissions) only give a partial view on
the firm’s CSP in the more general social and environmental meaning (Ioannou and Serafeim, 2012; Lydenberg et al., 1986). As a matter of fact, Waddock and Graves (1997) point at the requirement for a multidimensional measure of CSP that could be applied through a broad range of industries and large samples of firms. Until now, a lot of CSP measures have been utilized by scholars. Among others, these measures include survey tools (Aupperle et al., 1985), Kinder Lydenberg Domini, Inc. (KLD) (e.g., Boulouta, 2013; Graves and Waddock, 1994; Hafsi and Turgut, 2013; McWilliams and Siegel, 2000; Turban and Greening, 1996; Waddock and Graves, 1997), Sustainalytics (e.g., Surroca et al., 2013, 2010), and Thomson Reuters’ ASSET4 (ESG) (e.g., Cheng et al., 2014; Ioannou and Serafeim, 2012; Lys et al., 2014; Serafeim, 2014).
In this paper I utilize Thomson Reuters’ ASSET4 (ESG) dataset, as this dataset contains data on European firms as opposed to KLD, which is used by other scholars (e.g., Boulouta, 2013; Hafsi and Turgut, 2013). This dataset incorporates data on ESG dimensions of firms as of 2002. Researchers of Thomson Reuters’ ASSET4 (ESG) are particularly trained analysts who gather 900 evaluation points per firm. As stated by Thomson Reuters’ guiding principles, all the primary data used are unbiased and publicly accessible. Examples of data sources are, among others, CSP and annual reports, websites of NGOs, and various news sources. After collecting the data, the analysts transform the data into consistent components, which allows for quantitative inquiry. Then, the 900 evaluation points are utilized as inputs to an equally weighted framework to compute 250 key performance indicators (KPIs) and further bring those together in 18 categories within four pillars, being: (1) environmental performance pillar; (2) social performance pillar; (3) corporate governance pillar; (4) economic performance pillar. Each year, a firm obtains a z-score for every KPI, category, and pillar. By doing so, the performances of a firm are benchmarked with the other firms.
11
used energy, and recycled water, while social features naturally cover employee turnover, amount of training hours, charitable contributions, and controversies with regard to wellbeing and safety. Yet, the following question arises: how should one weigh these pillars? Previous scholars who use KLD encounter exactly the same problem. Until now, there is no theoretically originated ranking of prominence for the several stakeholders impacted by the environmental and social performance of firms as a directive for academic research. Hence, I follow the praxis of Waddock and Graves (1997) who use KLD to build a composite CSP index by giving equal weights, and hence equal importance, to both pillars. Similarly, Ioannou and Serafeim (2012) and Lys et al. (2014) construct a composite CSP index using the ASSET4 (ESG) dataset, giving equal weights to the environmental and social performance pillars. Thus, I formulate the CSP index as:
𝑌𝑖,𝑡 = CSP index (equally weighted average of the environmental and social performance
pillar scores),
where 𝑌𝑖,𝑡 is the CSP index for company i in year t.
3.3 BGD (independent variable)
In the academic literature, several measures of BGD are used. While some scholars use the number of females in the BoD (Carter et al., 2010) or a dichotomous variable which equals 1 if there is at least one woman in the BoD and 0 otherwise (Carter et al., 2003), most researchers use the percentage of females in the BoD as a measure for BGD (e.g., Adams and Ferreira, 2009; Boulouta, 2013; Campbell and Mínguez-Vera, 2008; Erhardt et al., 2003; Miller and Triana, 2009). To give a complete as possible view of the relation between BGD and CSP index, I therefore use all these three measures in my estimations. Hence, I formulate BGD as either:
𝑋1𝑖,𝑡 = Percentage of females (total number of females on the BoD divided by the total
number of BoD members), or:
12
𝑋1𝑖,𝑡 = Female dummy (dichotomous variable which equals 1 if there is at least one woman in the BoD and 0 otherwise),
where 𝑋1𝑖,𝑡 is the BGD for company i at the end of year t.
3.4 Control variables
According to Margolis and Walsh (2001), there are three control variables particularly important in investigating the relation between CSP and FP, namely industry effects, size, and risk. Hence, researchers who inquire the determining factors of CSP most often control for industry effects, size, risk and FP (Graves and Waddock, 1994; Ullmann, 1985; Waddock and Graves, 1997). Although important, I initially do not control for industry effects, as firm-fixed effects are reckoned to be more essential in explaining variability of results in CSP-related studies (Boulouta, 2013; López et al., 2007; Rumelt, 1991)4. Size is controlled for, as when firms grow and mature, they generally attract more attention from external stakeholders to whom they need to respond to more openly via social behaviors (Waddock and Graves, 1997). Riskier firms have a lower tendency to spend money on CSP-enhancing projects, as less capital could be dedicated to CSP-related issues that are not immediately connected to the economic survival (i.e., default prevention) of the firm (Orlitzky and Benjamin, 2001; Waddock and Graves, 1997). FP is an important control variable, as less profitable firms have less resources to use for CSP-related activities than more profitable firms (Campbell, 2007; Waddock and Graves, 1997). Next to these general controls, scholars stress the importance of including R&D intensity as control variable. R&D intensity and CSP are positively correlated, as aspects of CSP could come from product or process innovations (McWilliams and Siegel, 2000). To give an example: if a manufacturing company builds an innovative system to reduce CO2 emissions in its production process, this system positively affects the
environmental (i.e., CSP) outcomes of the company. Finally, scholars emphasize that BoDs with a higher ratio of outside BoD members could strengthen CSP, because they tend to show greater concern about CSP-related issues compared to inside BoD members (Ibrahim and
4
Variation in my data is not high enough to employ firm-fixed and industry-fixed effects simultaneously. To add robustness to my initial results, I do an estimation with industry-fixed instead of firm-fixed effects in section
13
Angelidis, 1995; Wang and Coffey, 1992). Therefore, it is obvious to include this control variable, board independence, too.
Operationally, the size of the firm is measured by the natural logarithm of the total number of employees at the end of year t for each firm i, as this is a proxy for the growth phase and maturity of the firm (Surroca et al., 2010; Waddock and Graves, 1997). Orlitzky and Benjamin (2001) show that several measures of risk, e.g., systematic risk (beta) and accounting risk (leverage), are employed in CSP-related studies. They argue it is a matter of choice, as both measures are operationalizations of the same underlying construct. In this paper, the risk of firm i is measured by the book value of long term debt divided by the book value of total assets at the end of year t, as the higher the debt level, the lower the possibility to use resources for CSP-related issues (Graves and Waddock, 1994; Waddock and Graves, 1997). Furthermore, I measure FP by the return on equity (ROE) for each firm i, which is the annual net income in year t divided by the book value of the average of t-1 and t year's common equity. The choice for ROE instead of return on assets (ROA), return on sales (ROS), or price to book value (PTBV) has to do with the fact that ROE is the most commonly used metric for FP in academic articles (Boulouta, 2013; Griffin and Mahon, 1997). Next, R&D intensity for each firm i is measured by the R&D expenses divided by total sales in year
t, following McWilliams and Siegel (2000). Lastly, board independence for each firm i is
controlled for by the total number of outside directors on the BoD divided by the total number of BoD members at the end of year t (Wang and Coffey, 1992). The five aforementioned control variables are represented as follows:
𝑋2𝑖,𝑡 = Size (natural logarithm of the total number of employees at the end of year t);
𝑋3𝑖,𝑡 = Risk (book value of long term debt divided by the book value of total assets at the end of year t);
𝑋4𝑖,𝑡 = ROE (annual net income in year t divided by the book value of the average of t-1 and t year's common equity);
𝑋5𝑖,𝑡 = Board independence (total number of outside directors on the BoD divided by the total
number of BoD members at the end of year t);
14
3.5 Moderators
To measure the moderating effects of national culture on the relation between BGD and CSP, I separately add Hofstede's (2001) MAS and IDV dimension variables as well as the interaction term with BGD to the econometric model. Hofstede's (2001) MAS and IDV scores measure on a scale from 0 to 100. They are measured per country and the organizations residing in that country accordingly:
𝑋7𝑖 = Hofstede’s MAS score of the country where company i resides; 𝑋8𝑖 = Hofstede’s IDV score of the country where company i resides;
𝑋9𝑖,𝑡 = (𝑋7𝑖 or 𝑋8𝑖) ∗ 𝑋1𝑖,𝑡 = moderating effect of either MAS or IDV with BGD.
3.6 Methodology
This methodology section consists of three subsections. The first subsection covers the estimation methods regarding Hypothesis 1, while the second one exhibits the estimation methods with respect to Hypotheses 2 and 3. The third subsection provides descriptive statistics and pair-wise correlation coefficients of the variables.
3.6.1 Estimation methods regarding Hypothesis 1
15
functions as a benchmark for the other estimates. Therefore, I estimate the following equation regarding Hypothesis 1:
𝑌𝑖,𝑡 = 𝛼0+ 𝛼1𝑋1𝑖,𝑡 + 𝛼2𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖,𝑡+ 𝑐𝑖 + 𝜐𝑡+ 𝜀𝑖,𝑡 , (1)
where i indexes the firms, t indexes years, 𝑌𝑖,𝑡 is the dependent variable, namely the CSP index (equally weighted rating of the environmental and social performance pillar scores), 𝛼0 is a constant, 𝛼1 and 𝛼2 are matching coefficients for BGD and all aforementioned control variables (size, risk, ROE, board independence, R&D intensity) respectively, 𝑐𝑖 is a firm-fixed effect, 𝜐𝑡 is a time-fixed effect, and 𝜀𝑖,𝑡 are the idiosyncratic errors or disturbances, as these change through i and t.
Nevertheless, it is reasonable to assume that another type of endogeneity, namely a reverse causality problem is also existing in the relation between BGD and CSP, as firms scoring higher on the CSP index could be more likely to raise BGD (Boulouta, 2013). If this is the case, panel data fixed-effects methods are unsuitable. The correct approach for this sort of endogeneity is the use of a 2SLS instrumental variable (IV) estimator (Brooks, 2008). To use this method, we need an instrument that is highly correlated with the endogenous explanatory variable, i.e., BGD, yet in essence uncorrelated with CSP index, except through control variables.
16
of female BoD members. Thus, I consider FFELFP as a good instrument for BGD. I formulate this IV as follows:
𝐼𝑁𝑆𝑇𝑅𝑈𝑀𝐸𝑁𝑇𝑖,𝑡 = FFELFP (number of full-time females divided by full-time employment [lagged by ten years]),
where 𝐼𝑁𝑆𝑇𝑅𝑈𝑀𝐸𝑁𝑇𝑖,𝑡 is the FFELFP – lagged by ten years - for company i at the end of year t.
In 2SLS estimations, one estimates two regression stages. The first stage is estimated to retrieve the fitted values of the endogenous explanatory variable, BGD, while the second stage estimates the impact of the fitted values of BGD on CSP index. Hence, in order to get the correct values of BGD to put in model (1), referred to as the second stage model in IV estimations (Brooks, 2008; Wooldridge, 2002), I first estimate the first stage via model (2). This model (2) estimates the fitted values of BGD using the instrument FFELFP and various control variables5:
𝑋1𝑖,𝑡 = 𝛽0 + 𝛽1𝐼𝑁𝑆𝑇𝑅𝑈𝑀𝐸𝑁𝑇𝑖,𝑡+ 𝛽2𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖,𝑡+ 𝑐𝑖 + 𝜐𝑡+ 𝜀𝑖,𝑡 , (2)
where i indexes the firms, t indexes years, 𝑋1𝑖,𝑡 is the endogenous variable BGD, 𝛽0 is a
constant, 𝛽1 and 𝛽2 are matching coefficients for the instrument and all aforementioned
control variables (size, risk, ROE, board independence, R&D intensity) respectively, 𝑐𝑖 is a firm-fixed effect, 𝜐𝑡 is a time-fixed effect, and 𝜀𝑖,𝑡 are the idiosyncratic errors or disturbances, as these change through t and i.
3.6.2 Estimation methods regarding Hypotheses 2 and 3
For Hypotheses 2 and 3 I use a similar 2SLS method as for Hypothesis 1 to investigate whether the contingent factors of national culture, i.e., MAS and IDV, have a significant moderating effect with BGD in the relation between BGD and CSP. Hence, I show whether and if so to what extent the cultural contingencies MAS and IDV play a role in the relation between BGD and CSP. The first stage is calculated in a similar vein as model (2). Subsequently, I estimate the second stage model (3), using the fitted values of the first stage:
5
17
𝑌𝑖,𝑡 = 𝛼0+ 𝛼1(𝑋7𝑖or 𝑋8𝑖) + 𝛼2𝑋9𝑖,𝑡+𝛼3𝑋1𝑖,𝑡+ 𝛼4𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖,𝑡+ 𝜐𝑡+ 𝜀𝑖,𝑡 (3)
Model (3) is similar to model (1) except for the inclusion of the cultural dimensions MAS or IDV, the interaction of MAS or IDV with BGD, and the exclusion of the firm-fixed effect. It is impossible to include a firm-fixed effect, since its inclusion would exclude the time-invariant cultural variables MAS and IDV.
3.6.3 Descriptive statistics and pair-wise correlation coefficients
I exhibit an overview of the descriptive statistics and pair-wise correlation coefficients for all variables in Table 1 and 2 respectively. To deal with possibly spurious outliers, I winsorize the variables ROE, R&D intensity, and risk at the 0.5 percent level on both sides. Table 1 shows levels of BGD that are higher than expected. While firms have on average 1.41 women in their BoD, the percentage of females and female dummy are 0.12 and 0.70 respectively. The average of 0.70 with respect to female dummy seems high, but is in line with the research by Carter et al. (2003). They find that 75.1 percent of Fortune 1000 firms have one or more women in their BoDs. With regard to the instrument FFELFP, variation is quite low with a standard deviation (SD) of 0.05. Considering the cultural dimensions of MAS and IDV, variation across the 18 countries in my sample is present. SDs of 21.59 and 11.93 for MAS and IDV correspondingly offer the opportunity to thoroughly examine the moderating effects of these dimensions.
18
Table 1
Descriptive statistics for board, firm, and country variables
Mean SD Min Max Observations
Board variables Percentage of females 0.12 0.12 0.00 0.67 5,292 Number of females 1.41 1.40 0.00 9.00 5,292 Female dummy 0.70 0.46 0.00 1.00 5,292 Board independence 0.54 0.26 0.00 1.00 4,755 Firm variables ROE 0.18 0.27 -0.76 2.30 6,624 R&D intensity 0.02 0.05 0.00 0.33 6,134 Size (ln) 9.43 1.88 0.00 13.38 6,518 Risk 0.19 0.16 0.00 0.78 6,853 CSP index 70.11 25.29 6.46 97.97 5,431 Country variables FFELFP 0.34 0.05 0.18 0.46 7,146 MAS 51.35 21.59 5.00 79.00 7,200 IDV 74.78 11.93 27.00 89.00 7,200 Table 2
Pair-wise correlation coefficients
1 2 3 4 5 6 7 8 9 10 11 12 1. CSP index 1.00 2. Percentage of females 0.17 1.00 3. Number of females 0.25 0.91 1.00 4. Female dummy 0.26 0.70 0.68 1.00 5. Board independence 0.14 0.15 0.09 0.14 1.00 6. ROE -0.05 0.02 -0.02 0.03 -0.02 1.00 7. R&D intensity 0.03 -0.06 -0.06 -0.04 0.03 -0.05 1.00 8. Size (ln) 0.51 0.10 0.20 0.14 0.10 -0.04 -0.01 1.00 9. Risk 0.02 -0.04 -0.05 -0.04 -0.03 0.07 -0.16 -0.02 1.00 10. FFELFP 0.05 0.24 0.27 0.18 -0.11 -0.03 -0.03 0.03 0.02 1.00 11. MAS -0.10 -0.37 -0.31 -0.18 -0.12 0.05 -0.02 -0.05 0.00 -0.25 1.00 12. IDV -0.06 -0.02 -0.07 0.02 0.08 0.14 0.02 -0.15 0.00 -0.22 0.35 1.00 4. Results
19
dimensions, using 2SLS estimations. In the third section, I provide four tests that improve the robustness of my results.
4.1 Results regarding Hypothesis 1
Initially, I test Hypothesis 1 over the period 2007 to 2014. Neither the OLS, nor the 2SLS estimations yield significant results. I show the results of the second stage IV estimations in Table 36. When I set eyes on the Wald F-statistics in Table 3, all columns show Wald F-statistics below ten. This indicates that the instrument FFELFP is not highly correlated with BGD7. This implies that the instrument is not a valid instrument for the three measures of BGD in the period 2007 to 2014. As a result, the fitted values of the three BGD measures from the first stage are imprecise. This means that the results of the second stage are not interpretable.
For this reason, I deliberately choose to only use a part of my sample from now on, namely the part covering the years 2011 until and including 2014. Several scholars maintain that the wake of the global financial crisis was the catalyst for a lot of firms to start improving their corporate governance practices (e.g., Ingley et al., 2011; Lauesen, 2013). One of these improved governance practices is more gender diversity in the BoD (Credit Suisse, 2012). However, BGD does not change from one day to another. To exemplify, male directors cannot immediately be replaced and it takes time to find suitable female candidates. Therefore, a higher female full-time employment ratio possibly did not directly lead to higher BGD (i.e., high correlation is absent) in the early years of the crisis. This makes FFELFP not a suitable instrument to employ over the period 2007 to 2014. Yet, in the period of the aftermath of the financial crisis until now, namely 2011 to 2014, I argue that FFELFP is possibly highly correlated with the three measures of BGD.
Table 4 exhibits the results of the OLS estimations for the period 2011 to 2014. On the one hand, column 1 and 3 show no significant effects of BGD on CSP index. On the other hand, the result from column 2 shows a significantly negative effect of BGD on CSP index at the one percent level (β=-0.370). This result suggests that BGD negatively impacts CSP index. For that reason, I reckon that this OLS coefficient is possibly downward biased due to reverse causality (i.e., it does not show the pure effect, which is expected to be positive).
6 OLS and first stage IV results are available at request by the author.
7 I perform Cragg-Donald Wald F-statistic tests to determine whether my instrument is highly correlated with the
20
21
Table 3
Second stage regressions: BGD and CSP in European firms in the period 2007 to 2014
This table displays the results of the second stage IV estimations with both firm-fixed as well as time-fixed effects. The European sample encompasses 18 countries and covers the period 2007 to 2014. Board-level data originate from Thomson Reuters’ ASSET4 (ESG). Firm-level data are retrieved from Worldscope and Thomson Reuters’s ASSET4. Country-level data are obtained from Euromonitor. Columns 1, 2, and 3 demonstrate results from CSP index regressed on the fitted values of percentage of females, number of females, and female dummy respectively. Board independence, size (ln), ROE, risk, and R&D intensity are the control variables. CSP index is the equally weighted average of the firm’s scores on the environmental and social performance pillars. Percentage of females, number of females, and female dummy are instrumented and are fitted values from the first stage IV estimations. Board independence is the total number of outside directors on the BoD divided by the total number of BoD members, size (ln) equals the natural logarithm of the total number of employees, ROE is annual net income in year t divided by the book value of the average of t-1 and t year’s common equity, risk is the book value of long term debt divided by the book value of total assets, and R&D intensity equals R&D expenses divided by total sales. All variables are measured at the end of year t. Asterisks indicate significance at 0.01 (***), 0.05 (**), and 0.10 (*) levels. Robust standard errors are shown in brackets under the coefficients.
22
Table 4
BGD and CSP in European firms in the period 2011 to 2014
23
Table 5
First stage regressions: identification of BGD
This table shows the results of the first stage IV estimations with both firm-fixed as well as time-fixed effects. The European sample encompasses 18 countries and covers the period 2011 to 2014. While board-level data originates from Thomson Reuters’ ASSET4 (ESG), firm-level data are retrieved from Worldscope. Country-level data are obtained from Euromonitor. Columns 1, 2, and 3 demonstrate results from the estimation of the fitted values of percentage of females, number of females, and female dummy regressed on an instrument. Board independence, size (ln), ROE, risk, and R&D intensity are the control variables. Percentage of females equals the total number of female members on the BoD divided by the total number of BoD members, number of females equals the total number of females on the BoD, and female dummy is a dichotomous variable that equals 1 if a firm has one or more female BoD members and 0 if not. FFELFP equals the total number of full-time females divided by full-time employment lagged by ten years within the country in which a firm has its headquarters. Board independence is the total number of outside directors on the BoD divided by the total number of BoD members, size (ln) equals the natural logarithm of the total number of employees, ROE is annual net income in year t divided by the book value of the average of t-1 and t year’s common equity, risk is the book value of long term debt divided by the book value of total assets, and R&D intensity equals R&D expenses divided by total sales. All variables are measured at the end of year t. Asterisks indicate significance at 0.01 (***), 0.05 (**), and 0.10 (*) levels. Robust standard errors are shown in brackets under the coefficients.
Independent variable 2011-2014
Percentage of females Number of females Female dummy
24
Table 6
Second stage regressions: BGD and CSP in European firms in the period 2011 to 2014
This table displays the results of the second stage IV estimations with both firm-fixed as well as time-fixed effects. The European sample encompasses 18 countries and covers the period 2011 to 2014. Board-level data originate from Thomson Reuters’ ASSET4 (ESG). Firm-level data are retrieved from Worldscope and Thomson Reuters’s ASSET4. Country-level data are obtained from Euromonitor. Columns 1, 2, and 3 demonstrate results from CSP index regressed on the fitted values of percentage of females, number of females, and female dummy respectively. Board independence, size (ln), ROE, risk, and R&D intensity are the control variables. CSP index is the equally weighted average of the firm’s scores on the environmental and social performance pillars. Percentage of females, number of females, and female dummy are instrumented and are fitted values from the first stage IV estimations. Board independence is the total number of outside directors on the BoD divided by the total number of BoD members, size (ln) equals the natural logarithm of the total number of employees, ROE is annual net income in year t divided by the book value of the average of t-1 and t year’s common equity, risk is the book value of long term debt divided by the book value of total assets, and R&D intensity equals R&D expenses divided by total sales. All variables are measured at the end of year t. Asterisks indicate significance at 0.01 (***), 0.05 (**), and 0.10 (*) levels. Robust standard errors are shown in brackets under the coefficients.
25
4.2 Results regarding Hypotheses 2 and 3
The results with respect to Hypothesis 1 show that the three measures of BGD positively impact CSP index in the period 2011 to 2014 if unobservable heterogeneity and reverse causality are accounted for. Although the results with respect to Hypothesis 1 refer to a general European conclusion on the relation between BGD and CSP, it is even more interesting to investigate whether and if so to what extent differences in national culture impact the relation between BGD and CSP, more specifically the MAS and IDV dimensions of Hofstede (2001). I separately estimate the moderating effects of MAS and IDV with BGD via IV regressions.
Table 7 shows the 2SLS regressions with the separate inclusions of Hofstede's (2001) MAS and IDV dimensions. Columns 1, 2, and 3 show the results of the inclusion of the MAS dimension variable, while columns 4, 5, and 6 exhibit the results of the inclusion of the IDV dimension variable. I include year dummies, but it is not possible to account for firm-fixed effects, as the cultural dimensions are time-invariant in my sample. Table 7 displays that the interaction term MAS*BGD is positively significant at the conventional levels if I use percentage of females and number of females as measures for BGD (β=0.733; β=0.030). Moreover, the coefficients for percentage of females and number of females are positively significant at the conventional levels (β=73.547; β=5.873). These results are in line with my expectations, as roles and inherent values of men and women are not the same in masculine countries (Hofstede, 2001). As a result, having more women in the BoD has more impact on the CSP of a firm in a masculine country compared to a firm in a feminine country. Yet, if I use female dummy as a measure for BGD, significance is absent although the coefficient is still positive (β=0.033).
With respect to the IDV dimension variable, the moderating effect with BGD is significant at all significance levels for the three measures of BGD (β=8.906; β=2.268; β=7.479), which is also in line with my expectations. Yet, all measures of BGD have negative coefficients at all significance levels (β=-518.630; β=-154.803; β=-501.816). A possible explanation for these significantly negative coefficients is multicollinearity between either BGD and IDV or BGD and IDV*BGD, as multicollinearity could cause coefficients to switch signs (Brooks, 2008). The correlation between BGD and IDV*BGD is indeed very high (0.982).
26
that it is not possible to include the interaction term IDV*BGD. Consequently, I create an IDV dummy variable to make a distinction between firms with high and low IDV scores. As the average IDV score in my sample is 74.78, I decide that the dummy equals 1 if a firm has an IDV score of 75.00 and 0 otherwise. Next, I run the same 2SLS regression as the one with respect to Hypothesis 1 except for the inclusion of the interaction term IDV dummy*BGD. I provide the results in Table 8. The interaction terms IDV dummy*BGD are insignificant, which shows that the positive effect of BGD on CSP is not stronger within firms residing in countries with higher IDV scores. These results lead me to reject my third hypothesis.
27
Table 7
BGD and CSP in European firms with Hofstede’s (2001) MAS and IDV dimensions in the period 2011 to 2014
This table displays the results of the second stage IV estimations with the separate inclusions of Hofstede’s (2001) MAS and IDV dimensions as well as time-fixed effects. The European sample encompasses 18 countries and covers the period 2011 to 2014. Board-level data originate from Thomson Reuters’ ASSET4 (ESG). Firm-level data are retrieved from Worldscope and Thomson Reuters’s ASSET4. Country-level data are obtained from Euromonitor and Hofstede (2001). Columns 1, 2, and 3 (Columns 4, 5, and 6) demonstrate results from CSP index regressed on the fitted values of BGD and MAS*BGD (BGD and IDV*BGD). BGD signifies either percentage of females, number of females, or female dummy. Board independence, size (ln), ROE, risk, and R&D intensity are the control variables. CSP index is the equally weighted average of the firm’s scores on the environmental and social performance pillars. MAS (IDV) are Hofstede’s (2001) masculinity (individualism) scores for the country in which a firm has its headquarters, both measuring on a scale from 0 to 100. Board independence is the total number of outside directors on the BoD divided by the total number of BoD members, size (ln) equals the natural logarithm of the total number of employees, ROE is annual net income in year t divided by the book value of the average of t-1 and t year’s common equity, risk is the book value of long term debt divided by the book value of total assets, and R&D intensity equals R&D expenses divided by total sales. All variables are measured at the end of year t. Asterisks indicate significance at 0.01 (***), 0.05 (**), and 0.10 (*) levels. Robust standard errors are shown in brackets under the coefficients.
28
Table 8
BGD and CSP in European firms with Hofstede’s (2001) IDV dimension variable: the dummy approach
This table displays the results of the second stage IV estimations with both firm-fixed as well as time-fixed effects. The European sample encompasses 18 countries and covers the period 2011 to 2014. Board-level data originate from Thomson Reuters’ ASSET4 (ESG). Firm-level data are retrieved from Worldscope and Thomson Reuters’s ASSET4. Country-level data are obtained from Euromonitor. Columns 1, 2, and 3 demonstrate results from CSP index regressed on the fitted values of percentage of females, number of females, and female dummy respectively. Board independence, size (ln), ROE, risk, and R&D intensity are the control variables. CSP index is the equally weighted average of the firm’s scores on the environmental and social performance pillars. Percentage of females, number of females, and female dummy are instrumented and are fitted values from the first stage IV estimations. IDV dummy is a variable that equals 1 if a firm has an IDV score of 75.00 and 0 otherwise. BGD signifies either percentage of females, number of females, or female dummy. Board independence is the total number of outside directors on the BoD divided by the total number of BoD members, size (ln) equals the natural logarithm of the total number of employees, ROE is annual net income in year t divided by the book value of the average of t-1 and t year’s common equity, risk is the book value of long term debt divided by the book value of total assets, and R&D intensity equals R&D expenses divided by total sales. All variables are measured at the end of year t. Asterisks indicate significance at 0.01 (***), 0.05 (**), and 0.10 (*) levels. Robust standard errors are shown in brackets under the coefficients.
29
4.3 Robustness tests
In this subsection I provide four robustness tests to strengthen my results. First, I employ the critical mass theory by Konrad and Kramer (2006) and Konrad et al. (2008) to see whether my results are robust for its possible implications. Subsequently, I control for industry effects instead of firm-fixed effects as they also possibly influence CSP. Then, I investigate whether my results remain significant if I use other metrics for FP, namely ROA, ROS, and PTBV. Lastly, I aim to make my results more robust by replicating my 2SLS estimations with a sample of S&P500 firms.
4.3.1 Critical mass theory
Recent academic research suggests that if there is not a critical mass of women on the BoD, the individual influences of women are minimal. Based on an interview study with 50 female directors of Fortune 1000 firms, Konrad and Kramer (2006) and Konrad et al. (2008) maintain that this critical mass is equal to three, which implies that a woman can only make a true contribution to BoD decision-making if she has at least two other women in her BoD. Recent empirical work even shows that BoDs with a minimum of three directors of each gender are at least 79 percent more active at BoD meetings than BoDs without such composition, which is primarily driven by female BoD members who are more involved when a critical mass of at least three females is reached (Schwartz-Ziv, 2015). Therefore, women in BoDs with less than three women (i.e., either zero, one, or two women) are considered to be tokens who are isolated and ignored. On the contrary, in BoDs with three or more women, females are not regarded as tokens but rather as influencers (Konrad and Kramer, 2006; Konrad et al., 2008; Torchia et al., 2011). Assuming that this theory is correct, it implies that one should only consider firms with at least three female BoD members when investigating the relation between BGD and CSP (Boulouta, 2013).
30
4.3.2 Industry effects
Although firm-specific factors are generally considered to be more important in explaining variability of results in CSP-related studies, differing industry dynamics could also have an influence in the relation between BGD and CSP (Boulouta, 2013; López et al., 2007; Rumelt, 1991). According to several scholars, different industries have different contexts with respect to e.g., impact on the environment, governmental regulations, enforcement processes, consumer-oriented nature, and stakeholder involvement (Griffin and Mahon, 1997; Hafsi and Turgut, 2013). An example is that manufacturing firms are expected to have a higher impact on the environment compared to service firms, since their business processes produce more emissions, consume more resources, and generate more waste (Hartmann and Uhlenbruck, 2015). Scholars Hafsi and Turgut (2013) acknowledge the existence of differing industry dynamics by employing a dummy variable which equals 1 when the firm is a manufacturing firm and 0 otherwise. Although this method is not incorrect, I prefer a more specific distinction of industries based on four-digit SIC codes, following Cheng et al. (2014) and Lys et al. (2014). Since there are nine industries in this classification, I create eight industry dummies to control for industry effects in the relation between BGD and CSP. I show the results in Table 10. For all three measures of BGD, the results show a positive relation with CSP at any significance level (β=84.268; β=6.480; β=41.927). Hence, the results of my inquiry remain the same when controlling for industry-fixed instead of firm-fixed effects.
4.3.3 Alternative measures for FP
In all previous estimations, I use ROE as a measure for FP. Despite the fact that ROE is the most commonly used metric for FP in academic articles (Boulouta, 2013; Griffin and Mahon, 1997), a significant minority of articles uses ROA, ROS, or PTBV as a measure for FP. Therefore, I re-estimate my model with these three alternative measures for FP to add robustness. The results in Table 11 strengthen my initial results, as the results remain significant at any significance level (β=43.529; β=43.728; β=37.730)8.
4.3.4 S&P500 sample
The studies by Boulouta (2013) and Hafsi and Turgut (2013) use a different ratings provider for CSP compared to my research, namely KLD instead of ASSET4. As CSP does
8 I only use percentage of females as a measure for BGD because of lack of space. The other two measures of
31
not presently have a commonly accepted definition, all definitions until now, including KLD’s and ASSET4’s, are not reckoned totally objective. Consequently, CSP ratings are subject to criticism on their capability to correctly predict CSP, i.e., research starts to question the construct validity of ratings (Chatterji et al., 2009; Semenova and Hassel, 2014). However, what could be even worse is low robustness across different CSP ratings providers, i.e., low convergent validity. Bouten et al. (2015) maintain that, based on interview findings, different CSP ratings providers have dissimilar theorizations of CSP and could therefore measure CSP in a different way. Therefore, it is questionable whether it makes sense to compare my results, based on ASSET4 and a sample of STOXX® Europe 600 firms, with the results of previous works, based on KLD and a sample of S&P500 firms. To circumvent the potential problem of low convergent validity, I test whether my results are robust by replicating my 2SLS estimations with a sample of S&P500 firms9. The inclusion of a time-fixed effect is not possible, since its inclusion would exclude the instrument FFELFP. This happens because of the fact that FFELFP has the same value for every firm in a particular year, as all firms reside in the same country. As I expect, the results in Table 12 demonstrate that there is significantly positive relation between BGD and CSP at any level of significance (β=382.045; β=35.891; β=281.255), also during the crisis period 2007 to 2010. Hence, this test improves the external validity of my results.
9
32
Table 9
BGD and CSP in European firms with at least three female BoD members in the period 2011 to 2014
This table displays the results of the second stage IV estimations with both firm-fixed as well as time-fixed effects. The European sample encompasses 18 countries and covers the period 2011 to 2014. Board-level data originate from Thomson Reuters’ ASSET4 (ESG). Firm-level data are retrieved from Worldscope and Thomson Reuters’s ASSET4. Country-level data are obtained from Euromonitor. Columns 1 and 2 demonstrate results from CSP index regressed on the fitted values of percentage of females and number of females respectively. Board independence, size (ln), ROE, risk, and R&D intensity are the control variables. CSP index is the equally weighted average of the firm’s scores on the environmental and social performance pillars. Percentage of females and number of females are instrumented and are fitted values from the first stage IV estimations. Board independence is the total number of outside directors on the BoD divided by the total number of BoD members, size (ln) equals the natural logarithm of the total number of employees, ROE is annual net income in year t divided by the book value of the average of t-1 and t year’s common equity, risk is the book value of long term debt divided by the book value of total assets, and R&D intensity equals R&D expenses divided by total sales. All variables are measured at the end of year t. Asterisks indicate significance at 0.01 (***), 0.05 (**), and 0.10 (*) levels. Robust standard errors are shown in brackets under the coefficients.
33
Table 10
BGD and CSP in European firms with industry dummies in the period 2011 to 2014
This table displays the results of the second stage IV estimations with time-fixed effects. Industry dummies are created based on the firms’ four-digit SIC codes following Cheng et al. (2014) and Lys et al. (2014). The European sample encompasses 18 countries and covers the period 2011 to 2014. Board-level data originate from Thomson Reuters’ ASSET4 (ESG). Firm-level data are retrieved from Worldscope and Thomson Reuters’s ASSET4. Country-level data are obtained from Euromonitor. Columns 1, 2, and 3 demonstrate results from CSP index regressed on the fitted values of percentage of females, number of females, and female dummy respectively. Board independence, size (ln), ROE, risk, and R&D intensity are the control variables. CSP index is the equally weighted average of the firm’s scores on the environmental and social performance pillars. Percentage of females, number of females, and female dummy are instrumented and are fitted values from the first stage IV estimations. Board independence is the total number of outside directors on the BoD divided by the total number of BoD members, size (ln) equals the natural logarithm of the total number of employees, ROE is annual net income in year t divided by the book value of the average of t-1 and t year’s common equity, risk is the book value of long term debt divided by the book value of total assets, and R&D intensity equals R&D expenses divided by total sales. All variables are measured at the end of year t. Asterisks indicate significance at 0.01 (***), 0.05 (**), and 0.10 (*) levels. Robust standard errors are shown in brackets under the coefficients.
Independent variable 2011-2014 CSP index (1) (2) (3) Percentage of females 84.268*** [11.188] Number of females 6.480*** [0.694] Female dummy 41.927*** [6.073] Board independence 1.098 6.763*** 2.997*** [1.398] [0.914] [0.658] Size (ln) 5.478*** 4.789*** 5.425*** [0.295] [0.382] [0.169] ROE -2.816*** -2.163** -3.700*** [1.064] [0.944] [1.004] Risk 8.868*** 8.109*** 9.641*** [2.411] [2.555] [2.987] R&D intensity 33.251*** 22.861*** 31.131** [3.206] [2.643] [12.428] Firm-fixed effects No No No
Industry dummies Yes Yes Yes
Number of observations 1,367 1,367 1,367
34
Table 11
BGD and CSP in European firms with alternative measures for FP in the period 2011 to 2014
This table displays the results of the second stage IV estimations with firm-fixed and time-fixed effects. The European sample encompasses 18 countries and covers the period 2011 to 2014. Board-level data originate from Thomson Reuters’ ASSET4 (ESG). Firm-level data are retrieved from Worldscope and Thomson Reuters’s ASSET4. Country-level data are obtained from Euromonitor. Columns 1, 2, and 3 demonstrate results from CSP index regressed on the fitted values of percentage of females. Board independence, size (ln), ROA, ROS, PTBV, risk, and R&D intensity are the control variables. CSP index is the equally weighted average of the firm’s scores on the environmental and social performance pillars. Percentage of females is instrumented and are fitted values from the first stage IV estimations. Board independence is the total number of outside directors on the BoD divided by the total number of BoD members, size (ln) equals the natural logarithm of the total number of employees, ROA is annual net income in year t divided by the book value of the average of t-1 and t year’s total assets, ROS is annual net income divided by total sales, PTBV is the market share price divided by the book value per share, risk is the book value of long term debt divided by the book value of total assets, and R&D intensity equals R&D expenses divided by total sales. All variables are measured at the end of year t. Asterisks indicate significance at 0.01 (***), 0.05 (**), and 0.10 (*) levels. Robust standard errors are shown in brackets under the coefficients.
35
Table 12
BGD and CSP in S&P500 firms in the period 2007 to 2014
This table displays the results of the second stage IV estimations with firm-fixed effects. The S&P500 sample encompasses 494 firms and covers the period 2007 to 2014. Board-level data originate from Thomson Reuters’ ASSET4 (ESG). Firm-level data are retrieved from Worldscope and Thomson Reuters’s ASSET4. Country-level data are obtained from Euromonitor. Columns 1, 2, and 3 demonstrate results from CSP index 2011-2014 regressed on the fitted values of percentage of females, number of females, and female dummy respectively. Board independence, size (ln), ROE, risk, and R&D intensity are the control variables. CSP index is the equally weighted average of the firm’s scores on the environmental and social performance pillars. Percentage of females, number of females, and female dummy are instrumented and are fitted values from the first stage IV estimations. Board independence is the total number of outside directors on the BoD divided by the total number of BoD members, size (ln) equals the natural logarithm of the total number of employees, ROE is annual net income in year t divided by the book value of the average of t-1 and t year’s common equity, risk is the book value of long term debt divided by the book value of total assets, and R&D intensity equals R&D expenses divided by total sales. All variables are measured at the end of year t. Asterisks indicate significance at 0.01 (***), 0.05 (**), and 0.10 (*) levels. Robust standard errors are shown in brackets under the coefficients.
