CONCENTRATION, AND REMUNERATION REPORT
WRITING STYLE: A STUDY OF READABILITY AND
TONE
by
JURGEN LIMBERG
University of Groningen
Faculty of Economics and Business
August 2012
Willem de Zwijgerstraat 31 bis
3583 HB Utrecht
(06)52677474
j.limberg@student.rug.nl
Student number: 1634399
ABSTRACT
This study examines the relationships between gender diversity on boards of directors and the degree of impression management applied to UK remuneration reports. In addition, the effects of ownership concentration on this relationship, as well on impression management in itself, are taken into consideration. After controlling for other firm and board characteristics, I find no significant association between board gender diversity and readability of remuneration reports. However, my results do show that higher ownership concentration is associated with poorer remuneration report readability. Additionally, neither board gender diversity nor ownership concentration are found to influence the optimistic tone, resulting from firm performance, in remuneration reports. My findings cast doubt on the belief that female board members improve the monitoring function of boards of directors. As for ownership concentration, the results support common theory on how ownership structure affects financial reporting quality.
Key words: Readability; Linguistic tone; Female board membership; Ownership concentration
Thesis supervisor: Dr. B. Qin
Acknowledgements: I would like to thank Bo Qin for all comments and ideas, as well as for the swift replies to all of my questions. Furthermore, I would like to thank Ineke de Jong, Yoong-Shin Teong and Angela van Teijen for collaborating on the data collection. I would also like to thank the
University of Groningen and PwC Utrecht for providing me with all necessary resources for conducting this research.
1 INTRODUCTION
Boards of directors, as well as top management, in modern-day corporations have always been dominated by male membership. Women trying to work their way up the corporate ladder have largely encountered, what is called, the glass ceiling (see e.g., Arfken, Bellar & Helms, 2004). Despite laws against gender discrimination in employment, like in the 1964 U.S. Civil Rights Act, typically few women are found in top corporate positions still. However, a recent regulatory development in Europe shows to be very effective in respect of gender diversity on corporate boards: Drastic measures were taken in Norway with a 2003 law requiring all public Norwegian firms to have at least 40 per cent of women on their boards of directors. By April 2008 all corresponding firms complied with this law (Adams & Ferreira, 2009). More European countries, like Iceland, Spain and Sweden, are also considering gender quotas like the one in Norway. In terms of gender equality, this is a giant leap in the right direction. Still, the question remains if such an exogenous shock to corporate boards is favorable to the performance of the firms in question. And in general, is higher board gender diversity likely to improve corporate governance and firm performance?
For the case of the Norwegian gender quota the obvious answer would be that firms, urging to meet the quota, appointed less qualified and less experienced (female) board members than they would have otherwise appointed. The Norwegian average percentage of female board members was only 6 per cent in 2002, and has risen to 40 per cent before April 2008. This means that, for the greater part, these women would be relatively inexperienced as board members. Ahern & Dittmar (2011) support this assumption by showing that the quota has caused poorer firm performance. Moreover, they state that the boards in question show signs of less experience, expressed in higher leverage and more acquisitions.
On the other hand several studies do indeed point out that the presence of a woman or women on a board of directors can be positively related to certain elements of corporate governance (Kesner, 1988; Bradshaw, Murray & Wolpin, 1996; Adams & Ferreira, 2009; Srinidhi, Gul & Tsui, 2011). Adams & Ferreira (2009) argue that women are more likely to take up an active monitoring role, such as joining a monitoring committee. They also note that independence is higher with female directors, because they are not part of the old boys club. Better monitoring behavior should prevent or correct unwanted managerial actions more effectively (see e.g., Xie, Davidson & DaDalt, 2003).
A significant example of unwanted managerial actions is impression management, which Leary & Kowalski (1990), in an influential paper, describe as an attempt to control other people’s impressions they form of you. These attempts to control impressions can be applicable to virtually any situation or process people are involved in. One example is corporate disclosure. In this perspective,
impression management can be seen as an attempt to control the way in which users interpret the information displayed in corporate reports. The most obvious reason to engage in impression management is to make things look better than they in fact are. For instance, prior research has shown that managers will try to conceal information they have to report in their mandatory disclosures, in order to obfuscate poor firm performance (Subramanian, Insley & Blackwell, 1993; Courtis, 1998; Li, 2008). Obfuscation is described by Adelberg (1979) as the situation in which managers attempt to conceal their failures in the accounting communication process. The author stresses that on the other hand managers will seek to emphasize their successes. The latter is also shown by Li (2010), who posits that managers refer to themselves more often in financial reports when their firm performs better.
The three mentioned studies on obfuscation of poor firm performance (Subramanian, Insley & Blackwell, 1993; Courtis, 1998; Li, 2008) show that readability of (parts of) corporate reports tends to be lower for poorly performing firms, when compared to that of firms with better performance. Other recent studies (Feldman, Govindaraj, Livnat & Segal, 2010; Huang, Teoh & Zhang, 2011; Rogers, Van Buskirk & Zechman, 2011) point out that managers alter the tone of the information in corporate disclosures. An abnormally optimistic tone in disclosures turns out to be associated with negative future firm performance (Huang et al., 2011) and high litigation risk (Rogers et al., 2011). This shows that management attempt to fool other parties, such as investors, into believing that the firm is in better health than it actually is. Feldman et al. (2010) and Huang et al. (2011) suggest that these veiling attempts are successful at first, but on the longer term investors and other parties will see through management’s reporting behavior. Huang et al. (2011) describe this market response as a “delayed reaction”. These examples show that both readability and tone of corporate reports can thus be seen as indicators of obfuscation.
It is thinkable that managers would want to prevent information on executive compensation and its determinants from becoming publicly available as well, particularly when executive compensation levels are to be considered disproportionate to firm performance. They could, therefore, engage in these forms of impression management in order to save face, with regard to compensation policies. This assumption should thus be reflected in companies’ compensation reports, which are a relatively new form of disclosure.
US Versus UK Compensation Reporting
In a 2005 paper by Jeffrey Gordon the advice goes out to the United States Securities and Exchange Commission (SEC) to implement new disclosure rules. This paper’s most significant recommendation concerns a requirement for proxy disclosure of executive compensation, which he calls the “Compensation Discussion & Analysis” (CD&A) (Gordon, 2005). This CD&A should “collect, itemize, and summarize all compensation elements for each senior executive, providing bottom line analysis and then a justification by the compensation committee of the compensation paid” (Gordon, 2005). The author posits that executive compensation is not in tune with executive performance, but is all the more a result of managers’ rent-seeking ability. This assumption is based on arguments raised by Bebchuk & Fried (2004) in their suggestively titled book Pay without
performance: The unfulfilled promise of executive compensation. Gordon (2005) argues that publicity
of compensation information, in the form of CD&A, would lead to a better bargaining position for compensation committees. The publicity would make it easier for them to withstand managerial pressure on compensation policy. Furthermore, it would evoke responses from shareholders and other stakeholders, which could also contribute to a better match between managerial pay and performance. The SEC gave ear to Gordon’s (2005) recommendation with the issuing of new disclosure rules in 2006 (http://www.sec.gov). These rules include mandatory reporting on executive compensation in the new Compensation Discussion & Analysis, in order to improve investors’ view on a company’s compensation policy.
Along with this CD&A requirement the SEC release (33-8732a) forces proxy statements, including CD&A, and other filings to be written in plain English. The SEC’s intention with this plain English requirement is to make reports and statements “easier to understand” (http://www.sec.gov). In scientific research, the ease of understanding is a very common definition of the word “readability” (see e.g., Gilliland, 1968; Loveland, Whatley, Ray & Reidy, 1973; Zibell, 2000). The most referred to author in this respect is George Klare, who defines readability as the ease of understanding due to the style of writing (Klare, 1963). From this I conclude that the plain English requirement is created to improve readability of disclosures filed by companies under SEC regulation. For CD&A it cannot be a matter of improving readability, since CD&A is a new disclosure requirement, but the SEC wants to ensure that these reports are easy to understand as well.
Recent research on impression management in CD&As was conducted by Laksmana, Tietz & Yang (2011). They find that in the first effective year of CD&A reporting, 2007, CD&A readability scores are significantly lower for companies with CEO pay higher than the CEO compensation
benchmark. This outcome proves the presence of obfuscation in CD&As, despite the plain English rule. Be that as it may, the authors attribute this obfuscation to poor enforcement of the rule. They indeed show that after stronger enforcement by the SEC the 2008 CD&As were, in general, easier to read.
Although Gordon’s (2005) CD&A proposal seems to have laid the groundwork for the 2006 SEC requirement, a very similar form of compensation reporting has been mandatory in the United Kingdom since 2002. UK remuneration reports cover almost exactly the same topics and items as CD&As do. This compromises the newness of US CD&A reporting. However, there is no plain English rule yet for UK financial reporting, which means that CD&As can be expected to be easier to read than UK remuneration reports. In other words: UK remuneration reporting is exposed to possible obfuscation to a greater extent than CD&A reporting is. With regard to optimistic or pessimistic tone in compensation reports, there are no indications to assume that the possibilities to engage in this form of impression management differ between UK and US reporting.
It is obvious that management obfuscation can be problematic for remuneration report users in terms of correctly interpreting the message the document contains. The better a company’s management conceals “unfair” executive compensation, the harder it will be for users, like investors, to judge the company’s compensation policy. This means that, particularly in the absence of a well-enforced plain English rule in the UK, there is need for a mechanism to control managers’ freedom to conjure with readability in remuneration reports in their own interest. In the same way, the possibilities for management to apply a misleading tone to remuneration reports need to be mitigated.
Research Question and Relevance
In this paper I will focus on female board membership as a possible instrument to control management obfuscation in remuneration reports. As several other studies show, female board members are likely to improve the monitoring function of the board (Adams & Ferreira, 2009) and other parts of corporate governance (Kesner, 1988; Bradshaw et al., 1996; Srinidhi et al., 2011). Therefore, I examine if higher female board representation leads to remuneration reports which are easier to read, compared to remuneration reports of companies with less or no women on their boards. Additionally, I inquire the possible relationship between female board representation and optimistic tone in remuneration reports.
Furthermore, I consider the impact of ownership concentration on both the readability and tone of remuneration reports. Leuz (2006) claims that differences in ownership concentration can affect
reporting incentives, and thus influence reporting behavior. This is also shown by Beuselinck & Manigart (2007), who find that companies with high ownership concentration exhibit lower financial reporting quality than companies with low ownership concentration. Hence, when readability and tone are viewed as aspects of reporting quality (see Merkl-Davies & Brennan, 2007), I expect ownership concentration to have a direct effect on the readability of remuneration reports. Similarly, I expect it to influence the optimistic tone of these reports, as explained further in Section 2. Moreover, I expect ownership concentration to affect the relationships between female board membership and the two indicators of possible obfuscation in remuneration reports, as theorized in the next section. This leads to the following research question:
To what extent are the readability and optimistic tone of remuneration reports influenced by female board representation, and to what extent does ownership concentration affect both this influence and the two writing style measures in themselves?
The importance of the tone and degree of readability of corporate reports lies in the shareholders’, other (potential) investing parties’ and analysts’ interpretation of the presented information. Their judgment fuels the investment decision, whether positively or negatively, and ultimately influences the market value of the firm in question (for a similar point of view, see Tan, Wang & Zhou, 2011). This is why research on impression management has to be conducted. Proof of firms deliberately making reports harder to read or applying tone management, for any reason of concealing the truth, can make investors and analysts more aware of possible concealment while analyzing reports. The awareness can drive them to dig deeper into the reports and enhance their judgment of a firm’s performance and health. It could also lead to a review of earlier investment decisions. Both of these examples should bring forth more accurate firm valuation.
Reports on director compensation are exceptionally important to analyze, because of their susceptibility to management obfuscation. As Gordon (2005) concludes, compensation of directors is not in line with their performance. This is not a message managers would like to communicate to stakeholders. Consequently, managers will try to mask this information in any possible way, one of which is making compensation reports harder to read. Another is to create a misleading report by applying an unusually positive tone. Again, proof of this obfuscation will aid stakeholders in their analysis of a firm’s health. Furthermore, compensation reports are a relatively new component of corporate reporting, hence opening a new field of research. UK remuneration reports have only been legislated since 2002, and US CD&As since 2006. Both forms of compensation reporting are aimed at
more transparency with regard to executive remuneration. Therefore, at this time, it is essential to find out if the desired transparency is indeed achieved. Content analysis of the reports is one of the tools to accomplish this.
As for readability research, UK remuneration reports are more suitable than US CD&As, since they are not subject to a plain English rule, whereas CD&As are. This leaves UK managers with more room to “play” with the readability of their reports than their US counterparts. Moreover, a ground-breaking development in UK regulation came shortly after the implementation of remuneration reporting. In 2002 say-on-pay regulations were submitted to UK Parliament. Say on pay means that shareholders can annually vote on executive compensation. These votes are non-binding, but still the voice of the shareholder is heard and firm reputation is at stake (Ferri & Maber, 2010). This should drive firms with a weak link between pay and performance to either improve this link or to be less transparent about executive compensation. The latter could influence the readability of remuneration reports, as well as the decision to apply tone management. This is a second reason to perform a study on UK firms instead of US firms, since a similar say-on-pay rule has not been effective in the US until last year.
My research contributes to existing literature on readability and linguistic tone of financial reports, as well as on their determinants. Although many prior studies have focused on readability of accounting reports (e.g., Barnett & Leoffler, 1979; Courtis, 1986; Smith & Taffler, 1992; Subramanian et al., 1993; Rutherford, 2003; Li, 2008), linguistic tone in corporate reporting (e.g., Demers & Vega, 2010; Feldman et al., 2010; Huang et al., 2011; Rogers et al., 2011) and various effects of gender diverse boards of directors (e.g., Kesner, 1988; Bradshaw et al., 1996; Burgess & Tharenou, 2002; Farrell & Hersch, 2005; Adams & Ferreira, 2009; Srinidhi et al., 2011), the latter has, to the best of my knowledge, not been linked to writing style or management obfuscation in any of those studies. This is where my research steps in to bridge the gap. The findings in this paper complement the mentioned studies on readability and tone of corporate reports by examining remuneration reports, specifically. My findings also provide more insight into what variables possibly determine obfuscation in financial reports.
Furthermore, this study contributes to the understanding of the effects of ownership concentration on corporate governance. Particularly, financial reporting quality is investigated. Confirming the results shown by Leuz (2006) and Beuselinck & Manigart (2007) would stress the fact that high ownership concentration can undermine the quality of financial reports. In addition, it would prove that ownership concentration impacts compensation reporting in particular.
For practitioners, the outcomes of this research are particularly interesting. If, in fact, the presence of women on the board of directors influences the quality of financial reports, measured by readability and optimistic tone, this has great implications for future board member appointment decisions. For example, if the relationship between female board membership and readability turns out to be significantly positive, this could lead to shareholders contending for more female board members in the fight against management obfuscation. Also, with the rise of gender quotas for boards of directors (see Srinidhi et al., 2011), the outcomes might be helpful to regulating institutions in making decisions in that respect. Moreover, my findings will, in conjunction with findings of prior research, aid financial report analysts in identifying possible obfuscation in compensation reports like reports on director remuneration. Recent research (Lehavy, Li & Merkley, 2011) has shown that less readable reports, in this case 10-K filings, press analysts to incur more effort in generating their own analytical reports. In addition, their forecasts prove to be less accurate for firms producing less readable reports. A similar problem arises from tone management in corporate reporting, in the light of impression management. The short-term market reaction to reports with an abnormally optimistic tone proves to be positive (Feldman et al., 2010; Huang et al., 2011). However, the market reaction will be more negative on the longer term. Huang et al. (2011) therefore conclude that an excessive positivity in the tone of a report disinforms the market.
Considering this, analysts can benefit from the results of studies of this kind, and improve their judgment of corporate reports. They can account for the likeliness of obfuscation, with reference to board characteristics like gender diversity. In a similar way, analysts can take into account the implications of ownership concentration for report readability and tone.
The remainder of this paper consists of four different sections. In the following section I will formulate the relevant hypotheses for this research, based on related theory. Next, a methodology section is provided in which the research methods are described in detail. Following that section the research results are displayed in Section 4. The final section gives a discussion of these results. Conclusions, as well as implications and recommendations for accounting practice and future research are also discussed in this section.
2 THEORY AND HYPOTHESES
General Agency Theory and Impression Management in Remuneration Reports
Ever since the separation of corporate ownership and control the debate continues around aligning the interests of directors and owners (for an overview, see Eisenhardt, 1989). In an influential
paper, Jensen & Meckling (1976) quote Adam Smith’s (1776) The wealth of nations, in which he beautifully describes the fundamental agency problem. My simplified interpretation is that directors of public companies cannot be expected to manage the firms’ funds as well as private companies’ directors would manage their funds. The plain reason for this phenomenon is that it simply is not their own money. This creates an environment in which directors are tempted to act in their own interest instead of that of the shareholders (see Fama, 1980). Differences in interests between managers and owners arise from the fact that decisions favorable to a manager are not by definition favorable to the company itself, and thus to its owners (for a similar point of view, see Edlin & Stiglitz, 1995). For example, whereas a manager could pursue short-term results in order to meet performance targets, shareholders are more concerned with longer term profitability. Healy & Palepu (2001) contend that this agency conflict is one of the two reasons for the demand for accounting disclosure. The other reason they give, that is information asymmetry, goes hand in hand with the agency conflict. Investors are not perfectly able to evaluate management decisions, because they do not have access to all relevant decision information the management uses. This could lead to inadequate market valuation of the company in question (see Healy & Palepu, 2001: 407-408). The authors describe the intended healing effect of regulated disclosure. This should provide outside investors with more information to evaluate management decisions more accurately. Remuneration reports are an example of this regulated disclosure in the UK.
However, managers still have some tricks up their sleeves. Although a vast amount of disclosure is mandated, as are remuneration reports, the style requirements leave preparers with at least some freedom in the way they present the required information. Prior studies have shown that management will abuse this freedom to conceal information they want to hold back from users of accounting disclosures, for example by applying an (unusually) optimistic tone (Sadique, In & Veeraraghavan, 2008; Feldman et al., 2010; Huang et al., 2011; Davis, Piger & Sedor, 2012), creating misleading graphs (Beattie & Jones, 1992; 2000; Courtis, 1997) or complex writing (Subramanian et al., 1993; Courtis, 1998; Li, 2008; Laksmana et al., 2011). These are forms of impression management, which, in their prominent psychological paper, Leary & Kowalski (1990) describe as “the process by which individuals attempt to control the impressions others form of them”. From a business perspective Beattie & Jones (2000) argue that management will try to create a self-serving presentation of corporate performance. Schrand & Walther (2000) call this “strategic disclosure behavior”. They state that management is more likely to exhibit this strategic behavior when performance is poorer than expected. Hence, for managers engaging in these forms of impression management, the goal is to distract disclosure users from information that could possibly leave a
negative impression. This makes remuneration reports exceptionally susceptible to impression management, because, for example, compensation policies disproportionate to firm performance could be unacceptable to shareholders and potential investors. In the case of UK firms this would not only lead to a bad reputation, but also has great implications for the say-on-pay votes of shareholders. If managers are able to set up remuneration reports in such way that users are less likely to find the disproportionality between pay and performance, the impression formed by these users should be less negative. Of course, this is an undesirable situation for disclosure users. It also raises the question how management behavior can be controlled, so that impression management in corporate reporting will be minimized.
Boards of directors are designed as a mechanism to control management behavior. Baysinger & Butler (1985) state that boards of directors serve to resolve the interest conflicts between a firm’s management and the owners. This is necessary, since the owners, or shareholders, are not able to perfectly monitor and control management behavior themselves. However, the presence of impression management (e.g., Beattie & Jones, 1992; 2000; Subramanian et al., 1993; Huang et al., 2011) in corporate reports proves that boards of directors are also fallible in resolving the agency conflict (see Fama, 1980). Therefore, possible ways of improving the governing function of boards of directors need to be explored. Gender diversity is one of those possibilities.
The Effects of Gender Diverse Boards of Directors
A 2003 Norwegian law required public firms to comply with a 40 per cent female board member quota. This law did not directly result from corporate governance considerations, but was intended as a means of improving equality in the corporate environment. However, the changes in Norwegian boards’ composition also led to changes in corporate governance, and ultimately firm performance. Ahern & Dittmar (2011) show that the quota has impacted firm performance. They find that, as a result of the gender quota, boards of directors overall show signs of less capable boards, like increased acquisitions and poorer operating performance. The reason the authors give is that these boards of directors have become younger with reference to average director’s age, which equals less experience. A logical explanation lies in the fact that firms suddenly had to recruit female board members en masse. Within six years the proportion of women on Norwegian boards rose from 6 to 40 per cent, as a result of the gender quota. Basically, this means that most of these recruited women had to be inexperienced as board members. The results of Ahern & Dittmar (2011) are in line with the findings of Adams & Ferreira (2009), who posit that gender diversity, on average, negatively affects
firm performance. However, the reason these authors give differs greatly from Ahern & Dittmar’s (2011) explanation. Whereas Ahern & Dittmar (2011) address poorer firm performance to a lack of experience with female board members, Adams & Ferreira (2009) state that gender diversity can result in too much monitoring. This is a result of the monitoring behavior of women, as described below. For well-governed firms, in particular, more monitoring proves to be counterproductive, with regard to firm performance. Hence, Adams & Ferreira (2009) conclude that mandated gender quotas for boards of directors should not be expected to cause increases in firm value.
On the other hand, this deterioration of firm performance does not necessarily mean that corporate governance as a whole suffers from gender diversity. As Adams & Ferreira (2009) point out, women directors exhibit better board attendance than men, and gender diversity has a positive influence on board attendance behavior of male directors. Furthermore, the authors show that women tend to join monitoring committees more often than men do, which suggests that gender diversity causes increases in monitoring effort. Xie et al. (2003) posit that better monitoring behavior should, in turn, prevent or correct unwanted managerial actions more effectively. One explanation for the enhanced monitoring behavior, exhibited by women, is that women are more independent than men are, as board members (Srinidhi et al., 2011: 1613). Adams & Ferreira (2009) state that this is a result of women not being part of the old boys club, often found in corporate boards. A second reason for improved monitoring by women is that women tend to exhibit less self-interest than men do. Arlow (1991) attributes this to women being “more ethical and socially responsible” than men, based on results of his business ethics survey among college students. Third, several studies suggest that women are less likely to tolerate opportunistic behavior (for an overview, see Srinidhi et al., 2011). Similar to the other two characteristics of women, this can be expected to improve monitoring behavior of boards of directors employing female members.
The increased monitoring with gender diverse boards could be a key element in answering the question of how management behavior can be controlled, in order to lower or minimize impression management in corporate reporting. Kesner (1988) describes results corresponding to Adams & Ferreira (2009). She finds that female board members are more likely to become committee members. Interestingly enough, this does not go for compensation committees. Adams & Ferreira’s (2009) research results are also supported by Srinidhi et al. (2011), who find that female board participation is positively related to earnings quality. The authors attribute this increased earnings quality to female board members improving the monitoring function of the board of directors, as suggested by Adams & Ferreira (2009). Similar results are found by Bradshaw et al. (1996). They find that, with higher proportions of women on a board of directors, board formalization goes up and the board meets more
frequently. Another interesting finding is that power sharing is relatively high throughout boards with high female representation. Bradshaw et al. (1996), however, cannot provide conclusive evidence to show whether this relationship is actually fueled by women’s merits, or that women deliberately choose these power-sharing boards.
Altogether, I have strong indications to assume that female board members, with their attitude towards monitoring, will have a positive influence on financial reporting quality, and thus remuneration report readability.
H1: Female board representation is positively associated with remuneration report readability.
As for tone management, a direct relationship between female board representation and tone positivity would not be logical to hypothesize. Opposite to readability, tone in itself is not a direct measure of financial reporting quality, since a more negative tone in reports would be appropriate in the case of poor performance, just as a more positive tone would be with strong performance. Therefore, the direction of any influence of female board membership on tone positivity cannot be reasonably assessed. For this reason I first take into account the impact of firm performance on tone positivity.
As I mentioned, pure logic would predict that strong firm performance will be reflected by positivity, or optimism, in corporate reports. On the other hand, poor performance should be reflected by a less positive tone in reports. However, this logic does not account for impression management. Schrand & Walther (2000) show that management will adapt their reporting behavior strategically, in order to conceal poor firm performance. More specific, their findings suggest that managers engage in reporting tactics to distract analysts from a negative earnings surprise. This suggestion finds support with many other studies on impression management. For example, studies show that managers use misleading graphs in reports (Beattie & Jones, 1992; 2000; Courtis, 1997), or increase textual complexity of reports (Subramanian et al., 1993; Courtis, 1998; Li, 2008; Laksmana et al., 2011) to camouflage poor performance.
Davis et al. (2012) state that positive and negative language substantially influence the processing of information. Specifically, they claim that an optimistic presentation of information, in terms of language, leads to more favorable evaluations than a pessimistic presentation does. This is an incentive for management to apply a more positive tone to their reports, particularly when performance is relatively poor. Proof of this form of tone management is scarce, as opposed to proof of obfuscation through complex writing. Still, recent studies show that firm performance and tone
positivity in firms’ reports are negatively associated. Huang et al. (2011) find that firms’ stock prices react positively to earnings press releases containing unusually optimistic language in the first instance. After a while this market reaction bends and turns negative, reflecting investors unveiling poor firm performance. The authors therefore conclude that the applied tone management disinforms investors about the firm’s performance. Feldman et al. (2010) report similar results for the management discussion and analysis (MD&A) sections of SEC filings, but only examine the shorter term effects. Nevertheless, their findings support the theory that managers use tone management for the purpose of making things look better than they are. Furthermore, Rogers et al. (2011) show that abnormally optimistic language in earnings announcements is positively related to the risk of securities lawsuits. Their results, again, point out that managers use tone management for the purpose of letting investors believe the firm performs better than it actually does.
In conclusion, I expect a negative relationship between firm performance and optimistic language. This relationship can be expected to apply to remuneration reports as well as it does to MD&As and earnings press releases. I assume this, since remuneration reports function as the justification of a firm’s compensation policy. This means that optimism in remuneration reports could be used to conceal any compensation decisions disproportionate to firm performance. Hence, poor firm performance could lead to unusually positive language in remuneration reports.
The positive effect of female board membership on corporate governance, described with H1, then is expected to affect the relationship between firm performance and optimism in remuneration reports. Since female board members will more closely monitor management behavior (Kesner, 1988; Bradshaw et al., 1996; Adams & Ferreira, 2009; Srinidhi et al., 2011), I expect this improved governance to diminish the tone management resulting from poor firm performance. In other words, female board representation moderates the negative relationship between firm performance and optimism in remuneration reports.
H2: The negative relationship between firm performance and optimism in remuneration reports is weakened by female board representation.
The Effects of Ownership Concentration
My main motivation to bring ownership concentration into relation with female board membership and impression management is that theory suggests that ownership concentration affects the impact of corporate governance mechanisms (e.g., Dhnadirek & Tang, 2003; Setia-Atmaja, 2009).
This has strong implications for the hypothesized effects (H1 and H2) of female board membership on financial reporting quality. The exact implications, as well as the rationale behind them, are clarified with my fifth hypothesis (H5) below.
A direct relationship between ownership concentration and financial reporting quality, however, should not be overlooked. Regarding this subject, I can rely on a wealth of literature, at least for the part of ownership concentration. Berle & Means (1932) are regarded by many (e.g., Demsetz & Lehn, 1985; Lskavyan & Spatareanu, 2006; Setia-Atmaja, 2009) as the instigators of this field of research. In their book, The modern corporation and private property, Berle & Means (1932) describe the dispersion of ownership in the U.S. and its implications for the concerning shareholders. They contend that dispersed owners become more and more powerless in controlling a company’s assets. In line with agency theory (Jensen & Meckling, 1976) the authors feel that this powerlessness leaves managers with more freedom to use a firm’s resources for their own benefit, leading to poorer firm performance. This means that more concentrated ownership could act as a monitoring device (Demsetz & Lehn, 1985). If so, it is probable that ownership concentration positively influences financial reporting quality. However, Beuselinck & Manigart (2007) find that companies with highly concentrated ownership produce lower quality accounting information than companies with more dispersed ownership. Their explanation for this outcome is that investors with low equity stakes in a company need higher quality financial reporting than investors with high equity stakes. This is because investors with high equity stakes assumedly have access to other ways of monitoring management behavior (also see Lehavy et al., 2011). The exact same view is presented by Leuz (2006), and Fan & Wong (2002) also find a negative relationship between ownership concentration and accounting information quality for Asian companies.
Accordingly, I expect remuneration reporting quality, measured by readability, to be negatively related to ownership concentration.
H3: Ownership concentration is negatively associated with remuneration report readability.
As theorized with H2, firm performance and optimism in remuneration reports are expected to be negatively associated, as a result of strategic reporting decisions of management (see Feldman et al., 2010; Huang et al., 2011; Rogers et al., 2011; Davis et al., 2012). The way in which ownership concentration affects this relationship should be very similar to the influence of ownership concentration on readability, since readability and tone are both measures of impression management (for a review of the literature, see Li, 2010).
With H3, above, I have explained that more concentrated ownership leads to lower quality financial reporting (Fan & Wong, 2002; Leuz, 2006; Beuselinck & Manigart, 2007). If holders of large stakes of a firm’s equity do not require a firm to provide high quality and highly informative reports, the incentives for these firms’ managers to engage in impression management are not mitigated. In fact, those managers are left with more freedom to strategically organize their reporting behavior (see Schrand & Walther, 2000). Of course, this is unfavorable to the more dispersed owners of such firms. These owners do not have access to the (other) monitoring devices concentrated owners do have access to (Berle & Means, 1932). At most, dispersed owners can free ride on some of the research (Shleifer & Vishny, 1986) and monitoring efforts of more concentrated owners. This suggests that dispersed owners will be the victims of poorer quality reporting, resulting from the presence of one or more large blockholders. Moreover, reports produced by firms with concentrated ownership will be less informative to potential investors as well. However, this could in turn be favorable to management, if the impression they leave in a report is more positive than the underlying firm fundamentals.
As a consequence of the additional room for impression management, created by concentrated ownership, tone management is likely to be increased in corporate reports. I therefore expect ownership concentration to have an intensifying effect on the theorized negative association between firm performance and optimism in remuneration reports.
H4: The negative relationship between firm performance and optimism in remuneration reports is intensified by ownership concentration.
Setia-Atmaja (2009) raises the argument that ownership concentration affects the impact of corporate governance mechanisms. Dhnadirek & Tang (2003) even state that ownership concentration can become a barrier for the functioning of other corporate governance mechanisms. I deliberately say “other”, because ownership concentration itself can function as a corporate governance mechanism as well (Demsetz & Lehn, 1985). For instance, if one investor holds a large share of a firm’s equity, he can be expected to have more say in the firm’s administration than more dispersed owners would have (Shleifer & Vishny, 1986). Ownership concentration could then form a barrier for the functioning of the board of directors, which is an example of (other) corporate governance mechanisms (Baysinger & Butler, 1985). For example, blockholders could press the board of directors to direct their attention towards ways of increasing profitability rather than to management’s reporting behavior.
Consequently, the relationship between female board representation and impression management would be affected.
Moreover, high quality of the information in corporate reports is less crucial for owners holding large equity stakes. They have access to other monitoring devices (Leuz, 2006; Beuselinck & Manigart, 2007), hence they require less informativeness from a firm’s reports. This could be taken into consideration by boards of directors, with regard to dividing their attention among different aspects of their function. The latter could lead to more application of impression management.
Additionally, Adams & Ferreira (2009: 292) find that the positive effects of female board representation on governance only apply to firms with weak shareholder rights. When shareholder rights are stronger, which could be compared with concentrated ownership, the effect of gender diversity is disadvantageous for the firm’s governance.
For these reasons I expect higher ownership concentration to moderate the positive relationship between female board membership and remuneration report readability.
H5: The positive relationship between female board representation and remuneration report readability is weakened by ownership concentration.
3 METHODOLOGY
Sampling Information
In order to test the hypotheses, data are obtained from a large data set constructed in collaboration with fellow students and supervisors over the past three years. The original data set contains 1,840 years for UK-listed firms for the financial years 2002 through 2009. Each firm-year consists of data on more than a hundred variables, such as executive remuneration features, board characteristics and details on firms’ equity. 2002 Was the year in which remuneration report regulation became effective in the UK, and 2009 is the last year for which data were processed into the data file at the start of this study. For consistency reasons, firm-years for firms with a financial year not ending on December 31 are grouped in the previous year if the reported financial year ends before June 30. For financial year ends on or after June 30, the year of the observation is considered to be the year itself. The reason to apply this rule is that some firms with, for instance, a financial year ending March 31, 2008 call this financial year 2008, while the larger part of the financial year lies in calendar year 2007.
The 1,840 initial firm-years comprise non-financial firms only. Financial firms were left out of the sample because their regulatory environment differs significantly from that of non-financial firms (Goodhart, Hartmann, Llewellyn, Rojas-Suárez & Weisbrod, 1998). These regulatory differences could influence reporting, board characteristics and other aspects relevant for this type of research. From the initial sample I have removed firm-years for which any of the needed data for the statistical analysis were missing. Note that for all 2009 observations necessary data were missing. These selection criteria have led to a final sample of 1,432 unique firm-year observations, representing exactly 252 different UK-listed firms.
In order to deal with outliers, I winsorize (Dixon, 1960) all continuous variables at the lower and upper one percentiles. Considering the size of the sample, this 98 per cent winsorization should be sufficient in neutralizing the noise in my empirical results caused by extreme values.
Measurement of Readability
To assess remuneration report readability I have used a formula approach, which is commonly used in relevant readability studies (e.g., Courtis, 1998; Li, 2008; Laksmana et al., 2011). This approach is much less time consuming than reading, or letting others read, each remuneration report and then grading these reports based on the ease of understanding (see Klare, 1963) of the report. Moreover, the formula approach is not subject to human factors, such as fatigue, intelligence or distraction, which could seriously affect the validity of any results. The evolution of information technology is another reason to turn to a computative approach of determining readability. Whereas before the rise of computer technology, readability scores had to be calculated manually, text passages can now be automatically processed into readability scores using (online) utilities. This neutralizes the time-consuming character (Li, 2010) of the formula approach with manual calculation. The main statistic I use to assess readability is the Flesch reading ease formula (FRE), published in 1948 by Rudolf Flesch. Of all readability statistics FRE is the most commonly used one in readability research (see Courtis, 1998), with regard to accounting practices. Many times it has been used in conjunction with other readability statistics, such as the Gunning Fog Index (e.g., Courtis, 1986; Smith & Taffler, 1988; Li, 2008), and sometimes as the sole measure of readability (e.g., Barnett & Leoffler, 1979; Courtis, 1998). The FRE score is calculated as follows:
Derived from this formula, longer sentences with more syllables per word generate lower FRE scores, which represent poorer readability. Table 1 presents general FRE scores and their corresponding level of readability. To illustrate, the first paragraph of the introduction of this paper has an FRE score of 48.87, which is just within the difficult reading category. Courtis (1986) points out that the length of words is associated with a reader’s speed of recognition, and the length of sentences is in turn associated with memory span. Accordingly, he states that word length and sentence length are sufficient predictors of readability. This encourages the use of a simple readability formula like FRE.
TABLE 1
Flesch Reading Ease rating
Description Rating Very Easy 100-90 Easy 90-80 Fairly Easy 80-70 Standard 70-60 Fairly Difficult 60-50 Difficult 50-30 Very Difficult 30-0
(adopted from Lewis, Parker, Pound & Sutcliffe, 1986)
To carry out a sensitivity analysis of the model I use the Gunning Fog Index (FOG), which has been designed by Robert Gunning in 1952. This test is used to estimate the level of education that is required to properly understand a text passage of any kind. As mentioned above, FOG has also been used as a readability measure in many prior readability studies of accounting reports (e.g., Courtis, 1986; Smith & Taffler, 1988; Gibson & Schroeder, 1994; Li, 2008). Calculation of the FOG index differs slightly from FRE calculation. Whereas the FRE formula uses the average of the syllable count of all words in the text passage, FOG only takes into account words with three or more syllables in that respect. The full FOG formula is as follows:
FOG = 0,4 ∗ # words ≥ 3 syllablesTotal # words ∗ 100 +Total # sentences൨Total # words
The FOG scores resulting from the formula represent a US grade level. This means that text passages with higher FOG scores require a higher educational level, with regard to understanding of the text
passage. For comparison, technical and scientific literature tend to have a FOG score of 17 or higher (Smith & Taffler, 1988) and even literature with scores of 14 and higher is considered to be ignored by readers, unless a reader has special motivation or interest (Courtis, 1986). Li (2008) implies that text passages with FOG scores higher than 18 are unreadable. He also states that a fog score between 12 and 14 would be optimal. The FOG score for the first paragraph of this paper is 14.11, which is very close to the optimal range. Interestingly enough, the first paragraph of Li’s (2008) introduction has a FOG score of 18.32. In his own terms, this would mean that this particular text passage of his own paper is unreadable.
In order to determine the scores for the remuneration reports of the sampled firms, digital versions of the firms’ annual reports were obtained from each firm’s website. For each observed year the text of the remuneration report was copied from the PDF files and pasted into Microsoft Word as unformatted text. In the interest of consistency in processing the text passages, all headings, bullets, tables, spacing and other layout content were removed, leaving only full sentences. Colons above tables and graphs have been replaced by points to mark those points as the end of a sentence. The edited passages have been processed into a readability score utility which can be found on the internet (http://www.online-utility.org). This tool generates multiple readability statistics, as well as a character, word and sentence count. Before students began gathering these data, thesis supervisors have compared the outcomes of manual calculation of readability scores to the scores generated by several (online) utilities, for the same text passages. The tool that has been used showed the least deviation from the manually calculated readability scores.
Measurement of Optimistic Tone
Similar to the measurement of readability, I use a formula approach in order to quantify tone positivity in remuneration reports. The same edited texts from firms’ remuneration reports, used for readability measurement, are processed using Diction software. This program calculates an optimism score (OptDict) based on the sum of positive words minus the sum of negative words (see e.g., Yuthas, Rogers & Dillard, 2002; Rogers et al., 2011; Davis et al., 2012). In formula, the Diction calculation looks as follows:
Optimism = [Praise + Satisfaction + Inspiration] − [Blame + Hardship + Denial]
found with the standard Diction method. For this sensitivity analysis, a different measure of optimistic tone, PosH, is selected. Henry (2008) uses an alternative list of positive words, which is designed to be more suitable for financial texts than the standard Diction word list. This alternative positive word list can be found on page 387 of her paper (Henry, 2008). Tone positivity is then determined by counting the frequencies of each listed positive word, using the alternative word list in Diction. The positivity statistic deducted from Henry (2008) is calculated by dividing the number of positive words by the total number of words in a remuneration report. The final PosH statistic is generated by multiplying this outcome by 100 per cent.
Independent Variables
Explanatory variables. Female board representation (GenderMix) is the main independent
variable for the primary relationship, described in the first hypothesis. I measure female board representation by the percentage of female supervisory board members. This is done for each firm-year combination. Hence, relatively more women on a board of directors would lead to a higher GenderMix figure. The statistical analysis for FRE, for example, should thus show a positive relationship between this figure and FRE score, since higher FRE scores reflect better readability. The information on female board membership is obtained from the annual report of a firm for the year in question. The observed number of female supervisory board members is then divided by the total number of supervisory board members to generate the final value needed for statistical analysis.1
With regard to the hypothesized effects in H3 through H5, ownership concentration is the main independent variable. This variable is organized into three categories:
- no shareholders holding five percent or more of a firm’s equity shares;
- one or more substantial shareholdings of between five and ten percent of the firm’s equity shares (Conc1);
- one or more substantial shareholdings of ten percent, or more, of the firm’s equity shares (Conc2). These values are also obtained from firms’ annual reports for the corresponding year. This information is mostly found under ‘substantial shareholdings’.
Dummies are created for each category of ownership concentration. In case of a substantial shareholding of ten percent or more, shareholdings of between five and ten percent are ignored. I have selected the category of no substantial shareholdings as the holdout group in the statistical analysis.
1
Additional tests using an alternative measure of GenderMix have been performed to test if the theorized effects of board gender diversity on the dependent variables kick in at a certain point corresponding with high female board representation in the sample. In order to perform this test, dummies were created for top decile GenderMix statistics. The alternative measure did not change the regression analysis results qualitatively.
As for the hypotheses concerning optimism in remuneration reports (H2 and H4), a third explanatory variable is in place. These hypotheses contain the element of a negative relationship between firm performance and optimism. Huang et al. (2011) use return on assets as the measure of firm performance in their study on management tone. However, based on the market reactions to the use of positive tone (Feldman et al., 2010; Huang et al., 2011), it would be more appropriate to implement a market-based measure of firm performance. Therefore I use Tobin’s Q to assess firm performance (Adams & Ferreira, 2009). The impacts of ownership concentration and female board representation on the relationship between performance and optimism will then be determined using interaction terms.
Control variables. Aside from the main independent variables, nine control variables that are
either known or expected to influence the dependent variables are included in the statistical model. First, my results are controlled for firm size (Li, 2008; Srinidhi et al., 2011; Laksmana et al., 2012), measured by market value for each firm-year, as per year end. Laksmana et al. (2012) claim that sensitivity to political costs could increase with size, which would lead to more readable compensation reports. A similar rationale could be applied to tone positivity, as to readability. However, the complexity of compensation policy is higher with larger firms, hence leading to less readable compensation reports. Laksmana et al. (2012) therefore do not predict a sign for the impact of size on obfuscation. To avoid heteroscedasticity, as well as skewness (see Gu, Lee & Rosett, 2005), within the model I take the natural logarithm of all market value figures (LnMV).
Second, the use of Long Term Incentive Plans (LTIPs) is accounted for in the regression analyses. This variable (LnLTIP) is measured as the natural logarithm of total LTIP compensation plus 1 for each firm-year. One reason to control for LTIP use is that De Jong (2012) finds a negative association between the use of LTIPs and readability of remuneration reports. She posits that complicated descriptions of performance targets associated with LTIPs are the cause of this negative influence on readability. In addition, she finds that the use of LTIPs is positively associated with the optimistic tone in remuneration reports. Therefore, inappropriate regression estimations due to noise caused by differences in use of LTIPs among the sample need to be ruled out.
Third, I take board independence into account. Gupta & Fields (2009) state that board independence is believed to be significant and effective as a corporate governance mechanism, just as I expect female board membership to be. By controlling for this variable I make sure that board independence is not overlooked as a possible source of interference with the results of my analysis.
Board independence (Independ) is measured as the percentage of independent directors on the board of directors, based on the information in each year’s annual report.
Fourth, my results are controlled for average supervisory board member tenure (Tenure). On the one hand, longer periods of board membership could compromise a member’s independence, leading to poorer corporate governance. On the other hand, longer tenure means more knowledge of the company (Srinidhi et al., 2011) and better connections with the firm’s management, which could lead to better corporate governance. Either way, board members’ tenure could disturb the relationship between female board membership and obfuscation, and should thus be controlled for. Board member tenure is measured as the average number of years on the firm’s board of directors per board member. Tenure data have been obtained from BoardEx for each firm-year combination.
Fifth, I control for executive directors being members of the remuneration committee, next to their directorship. This form of duality means that an executive director takes part in the decision-making process with regard to his own compensation. Fundamentally, this is an undesirable situation that could catalyze obfuscation. The variable (ExecReCom) is measured as the percentage of remuneration committee members occupied as executive directors in the same company.
The sixth control concerns the presence of a compensation consultant (Consult), which is a dummy variable. If the firm makes use of one or more compensation consultants, this should lead to a more balanced remuneration package. Hence, it possibly diminishes obfuscation, since executive compensation reflects performance more accurately.
Seventh, the results are controlled for overpayment of CEOs. This is a variable expected to influence obfuscation in compensation reports (Laksmana et al., 2012), as explained in Section 2. Overpay is a dummy variable that equals 1 if the deviation between the CEO’s actual compensation and the expected compensation lies in the top quartile of the initial sample, and 0 otherwise. The final two control variables are industry and year dummies, respectively to account for industry and year fixed effects (see e.g., Clatworthy & Jones, 2003; Lel & Miller, 2008). Fama & French’s (1997) 12 industries classification is used to assess industry dummies. As for year dummies, 2002 is the holdout group.
In the regression analysis for readability, Tobin’s Q serves as a tenth control variable instead of an explanatory variable, as with optimistic tone. Concerning readability, Tobin’s Q is not regarded as an explanatory variable, since firm performance is not part of the hypothesized effects in H1, H3, and H5. However, firm performance is expected to influence readability (Subramanian et al., 1993; Courtis, 1998; Li, 2008), and thus needs to be controlled for.
Statistical Models
Readability. The following model is the full regression model used for FRE, based on the
explanatory and control variables described above.
FRE = β0 + β1 GenderMix + β2 Conc1 + β3 Conc2 + β4 LnMV + β5 LnLTIP + β6 Tobin’s Q + β7 Independ + β8 Tenure + β9 ExecReCom + β10 Consult + β11 Overpay + ∑ βj Industry dummies + ∑ βk Year dummies + εᵢ
In this model ‘βᵢ’ are the estimated coefficients for each of the variables in this model. These coefficients are estimated using ordinary least squares regression. With regard to the sensitivity analysis of the model, the right-hand side of the equation does not change. The one distinction is that FOG is used instead of FRE as the dependent variable. Furthermore, ‘εᵢ’ represents the error term.
Optimistic tone. The models used for regression analysis concerning optimism, using Diction
optimism scores, are as follows:
OptDict = β0 + β1 Tobin’s Q + β2 GenderMix + β3 GenderMix*Tobin’s Q + β4 LnMV + β5 LnLTIP + β6 Independ + β7 Tenure + β8 ExecReCom + β9 Consult + β10 Overpay + ∑ βj Industry dummies + ∑ βk Year dummies + εᵢ
and
OptDict = β0 + β1 Tobin’s Q + β2 Conc1 + β3 Conc2 + β4 Conc1*Tobin’s Q + β5 Conc2*Tobin’s Q + β6 LnMV + β7 LnLTIP + β8 Independ + β9 Tenure + β10 ExecReCom + β11 Consult + β12 Overpay + ∑ βj Industry dummies + ∑ βk Year dummies + εᵢ
In these models, ‘GenderMix*Tobin’s Q’ denotes the interaction term for the effect of female board representation on the relationship between firm performance and optimism in remuneration reports. ‘Conc1*Tobin’s Q’ and ‘Conc2*Tobin’s Q’ denote the interaction terms for the effect of ownership concentration on the relationship between firm performance and optimism in remuneration reports. All other variables and coefficients denote the same as in the readability model, shown above.
With regard to the sensitivity analyses of these models, the right-hand sides of the equations do not change. The only distinction is that Henry’s (2008) positivity score (PosH) is used instead of OptDict as the dependent variable.
4 RESULTS
Descriptive Statistics
Table 2 presents descriptive statistics for all dependent and independent variables. Interestingly enough, the mean for GenderMix is 7.32, meaning that the average board of directors consists of almost 93 per cent male members. Moreover, the median figure for GenderMix (.00) shows that most firms have no women on their boards of directors whatsoever. On itself, this strongly supports the assumption of the presence of the glass ceiling I mentioned in the introduction. Introducing a gender quota, like the one in Norway, would thus cause a huge shift in board composition. There are, however, some UK boards that would already meet a 40 per cent gender quota, as can be derived from the 99thpercentile figure for GenderMix (42.90).
The readability statistics, FRE and FOG, show that remuneration reports are typically difficult to read. When FRE statistics in Table 2 are categorized (see Table 1), they indicate that more than 25 per cent of the reports in the sample are in fact very difficult to read. The FRE’s mean (32.46) and median (32.67) of the sample are also much closer to the ‘very difficult’ than to the ‘fairly difficult’ category. As for the FOG measure, the 75th percentile figure (18.34) reveals that more than a quarter of the sampled reports is unreadable, in terms of Li’s (2008) categorization. The FOG score average (17.38), with a standard deviation of only 1.54, indicates that remuneration reports are comparable with technical and scientific literature, regarding readability. The results show strong consistency between the two different readability measures. Nevertheless, the first percentile figure for FOG (12.93) tells us that some of the reports are within the range of optimal readability, while the 99th percentile figure for FRE (46.93) still denotes difficult reading.
As for optimism, the OptDict scores appear to be quite normally distributed among the sample. Surprisingly, the mean (.32) and median (.20) figures are very close to zero optimism, according to Diction. This would mean that the tone in an average remuneration report is neither optimistic nor pessimistic. The other measure of optimistic tone, PosH, ranges from .46 per cent positive words at the first percentile to 2.48 per cent at the 99th percentile. Although the percentage difference looks small, this indicates the use of only one positive word in every 217 words for the first percentile to one positive word in every 40 words for the 99th percentile. In other words, there are more than five
times as many positive words in remuneration reports at the 99thpercentile, compared to reports at the first percentile.
The mean for dummy variable Conc2 (.55) points out that around 55 per cent of the sampled firms have at least one blockholder that holds 10 per cent or more of the firms’ equity. The Conc1 mean (.30) shows that, if there is no 10 per cent blockholder, two out of every three of these firms do have a substantial shareholder of between 5 and 10 per cent. In other words, only 15 per cent of the sampled firms have no blockholders of 5 per cent or more of the firms’ equity.
Other results worth mentioning are those for the control variables board independence, the use of compensation consultants, and tenure. Independ exhibits a surprisingly even distribution among the sample, with a mean of 49.47 per cent of independent directors on the board. This result shows that, on average, half of the members of any board of directors are not independent. Combined with the percentile figures, this points out that board independence differs among UK-listed firms in a considerable manner. The Consult statistics in Table 2 show that by far most of the sampled firms make use of a compensation consultant. The mean for Consult (.82), which is a dummy variable, points out that 82 per cent of UK-listed firms hire a compensation consultant. Finally, the Tenure statistics tell us that the average supervisory board member serves on the same board for just under five years (4.79).
TABLE 2 Descriptive statistics
Variable Mean Median Std. Dev 1st 25th 75th 99th N
FRE 32.46 32.67 5.62 14.76 29.28 35.94 46.93 1,432 FOG 17.38 17.48 1.50 12.93 16.51 18.34 20.66 1,432 OptDict .32 .20 3.32 -8.38 -1.68 2.28 10.03 1,432 PosH 1.35 1.31 .39 .46 1.08 1.60 2.48 1,432 GenderMix 7.32 .00 11.50 .00 .00 14.30 42.90 1,432 Conc1 .30 .00 .46 .00 .00 1.00 1.00 1,432 Conc2 .55 1.00 .50 .00 .00 1.00 1.00 1,432 Tobin's Q 1.09 .97 .48 .51 .86 1.14 3.27 1,432 LnMV 6.75 6.60 1.64 3.45 5.58 7.66 11.54 1,432 LnLTIP 10.42 12.76 5.48 .00 11.27 13.75 16.11 1,432 Independ 49.47 50.00 13.66 13.09 40.00 60.00 80.00 1,432 Tenure 4.79 4.40 2.56 .43 3.10 5.90 14.21 1,432 ExecReCom 30.40 25.00 32.75 .00 .00 50.00 100.00 1,432 Consult .82 1.00 .39 .00 1.00 1.00 1.00 1,432 Overpay .26 .00 .44 .00 .00 1.00 1.00 1,432
The correlations among all variables used in the regression analysis are given in the correlation matrix (Table 3) on page 28. The values in this table denote Pearson’s correlation coefficients.
The correlations between GenderMix and both FRE and FOG are -.049 and .030, respectively, with p-values exceeding .05. GenderMix does significantly correlate with both optimistic tone measures, OptDict (r = .081, p<.01) and PosH (r = .160, p<.01). As for ownership concentration dummies and dependent variables, the correlations between both Conc1 and Conc2, and FOG are .087 (p<.01) and -.095 (p<.01), respectively. For FOG instead of FRE these are -.104 (p<.01) and .096 (p<.01), respectively. I find insignificant correlations of -.023 and .008 between both Conc1 and Conc2, respectively, and OptDict. Finally, Conc1 and Conc2 correlations with PosH are .042 (p>.05) and -.075 (p<.01), respectively. The correlations between Tobin’s Q and the dependent variables, FRE, FOG, OptDict, and PosH are .029, -.026, .046, and .064. The correlation between Tobin’s Q and PosH is the only significant one out of those four (p<.05). None of these correlations indicate a multicollinearity issue, since their coefficients are amply below the arbitrary cut-off point of r = .8 (Farrar & Glauber, 1967).
As I expected, based on the descriptive statistics, FRE and FOG strongly correlate with a negative sign (r = -.806, p<.01). Furthermore, this strong negative correlation could have been predicted as a result of the characteristics of the FOG and FRE formulas as well. Both measures use a syllable count and account for sentence length as a basis. The negative sign results from the way in which the formulas are designed (see Section 3). Since these dependent variables are not plotted in the same regression analysis, multicollinearity as a result of this correlation cannot occur.
The correlation between OptDict and PosH is significant as well, though the correlation coefficient (r = .225, p<.01) is not as high as between FRE and FOG. Obviously, the sign of this coefficient is positive, since both OptDict and PosH are measures of positivity. Use of different word lists for OptDict and PosH, and not taking negative words into account for PosH are the most logical explanations for the relatively low correlation coefficient between the two tone positivity measures. Another high correlation value (r = -.723, p<.01) appears between Conc1 and Conc2. Again, this is not surprising. Derived from the mean statistics in Table 2, I conclude that 85 per cent (.30+.55) of the total number of hits on ownership concentration dummies are accounted for by Conc1 and Conc2. This means that the holdout group of no blockholders accounts for only 15 per cent. Consequently, if the observation on ownership concentration for a specific firm-year is not categorized as Conc2, the probability of a categorization as Conc1 is very high. Hence, the correlation between Conc1 and Conc2 is negative, as well as it is very high.
TABLE 3 Correlation matrix
Variable FRE FOG OptDict PosH
Gen-
der-Mix Conc1 Conc2
Tobin'
s Q LnMV
Ln-LTIP Independ Tenure
Exec- Re-Com Consult FRE FOG -.806** OptDict .006 .005 PosH .069** -.091** .225** GenderMix -.049 .030 .081** .160** Conc1 .087** -.104** -.023 .042 -.010 Conc2 -.095** .096** .008 -.075** -.061* -.723** Tobin's Q .029 -.026 .046 .064* -.013 .025 -.006 LnMV -.088** .079** .065* .073** .295** -.001 -.141** .093** LnLTIP .002 .037 .107** .217** .131** .024 -.090** .012 .351** Independ -.132** .101** .024 .069** .166** -.004 .002 -.035 .319** .250** Tenure .079** -.094** .025 .053* -.100** .067* -.042 -.035 -.107** -.071** -.307** ExecReCom .004 .017 .040 -.034 .099** -.046 -.033 .005 .122** .084** .112** .000 Consult -.091** .149** 0.33 .123** .086** -.081** .036 -.037 .261** .310** .239** -.121** .022 Overpay -.047 .067* .078** .051 .016 -.060* .078** -.044 .039 .240** .124** -.081** .003 .079** ¹ N = 1,432
* significant correlation at the 95% confidence level (two-tailed) ** significant correlation at the 99% confidence level (two-tailed)
Independ and Tenure are negatively correlated (r = -.307, p<.01), which indicates that longer tenure of supervisory board members diminishes board independence. Obviously, this is in line with the existing theory.
Furthermore, interesting correlations are found between firm size (LnMV) and both Independ and GenderMix. The coefficients for these correlations imply that the percentage of female board members increases with firm size (r = .295, p<.01), and that independence positively correlates with firm size even more (r = .319, p<.01).
Finally, LnLTIP shows strong correlations with both LnMV and Consult. Whereas the first association indicates that larger firms use more LTIPs in director compensation, the association between LnLTIP and Consult indicates that more LTIPs are used when a firm hires a compensation consultant. This is in line with Conyon, Peck & Sadler (2009).
Regression Analyses and Tests of Hypotheses
Readability. The results of the multiple linear regression analysis of the statistical model for
FRE are presented in Table 4. Based on these results I will draw my conclusions with regard to the hypotheses concerning readability of remuneration reports (H1, H3, and H5). Model 1 contains control variables only, in order to be able to measure adjusted R² change as a result of adding the explanatory variables. Models 2a, 2b and 2c are used to test for the hypothesized main effects on readability, described in H1 and H3. Whereas Model 2a contains all control variables and GenderMix as the explanatory variable, model 2b accounts for ownership concentration (Conc1 and Conc2) instead of GenderMix. Model 2c represents the full statistical model for FRE, as described in Section 3.
For the purpose of testing for the hypothesized effect of ownership concentration on the theorized relationship between GenderMix and remuneration report readability (H5), interaction terms are added to the full model. Model 3 is the model including these interaction terms.
The adjusted R² values in Table 4 show that introducing GenderMix (Model 2a) to the regression does not cause any change in adjusted R² (F = 3.277, p<.01). This indicates that there is no noteworthy shift in the model’s variance as a result of adding GenderMix. However, when ownership concentration dummies are introduced to the regression (Model 2b) instead of GenderMix, a substantial variance shift can be observed. Adjusted R² from Model 1 to Model 2b changes from .041 to .051, which is a percentage change of 24.4 (F = 3.782, p<.01). The effect of ownership concentration on the model’s variance is not offset or intensified by GenderMix in the full model,
