Fighting Hypocrisy: External Assurance and
CSR Decoupling
Master Thesis MSc Accountancy
Willem Westhuis S2954869 w.g.westhuis@student.rug.nl University of Groningen June 22, 2020 Supervisor: dr. N. Hussain Word Count: 10,055
Abstract
This study investigates whether external assurance and its quality matter in reducing the decoupling of CSR disclosure and performance. Such practices are hypocritical and can have adverse consequences for stakeholders. Considering decoupling an information asymmetry problem, it is argued that external assurance is an efficient response where voluntary disclosure alone falls short (H1). Subsequently, it is argued that assurance by Big 4 accounting firms is of higher quality (H2). To test these hypotheses, international archival data is used. Decoupling is calculated as the absolute difference of disclosure and performance scores, provided by the Thomson Reuters ASSET4 and Bloomberg databases. For 13,555 firm-years of the total sample it is indicated whether the report was assured or not. Findings support both hypotheses, although additional analyses show different results for understating and overstating firms. They add to the limited literature on CSR decoupling by demonstrating that both assurance and its quality are significant factors in reducing decoupling. Furthermore, it offers interesting insights in the field of auditing as it assesses the effectiveness of CSR assurance and quality differences between accounting and non-accounting providers. In addition, findings are helpful for investors questioning the effects of assurance when faced with decoupling practices.
Table of Contents
1. Introduction ... 4
2. Theory and Hypothesis Development ... 6
2.1 CSR Performance ... 6
2.2 CSR Disclosure ... 7
2.3 CSR Decoupling ... 8
2.4 Hypothesis Development ... 13
3. Methodology ... 17
4.1 Sample and Data Sources ... 17
4.2 Variables and Measures ... 17
4.3 Research Model ... 20 4. Results ... 21 5. Discussion ... 30 6. Conclusion ... 33 References ... 36 Appendices ... 41
1. Introduction
Reporting on Corporate Social Responsibility (CSR) activities is an increasingly popular way for firms to show that they feel publicly accountable for their social and environmental record (Hemingway & Maclagan, 2004). However, to fulfill its purpose of transparency towards stakeholders instead of being a hypocritical attempt to improve corporate image, it is crucial that disclosure of CSR performance is reliable and correct (Boiral, 2013; Sauerwald & Su, 2019). Nevertheless, literature has found that CSR disclosure rarely reflects actual performance (Pope & Wæraas, 2016). In an extreme but telling example, the environmental disaster caused by explosion of the Deepwater Horizon platform followed only a few months after BP, the operating oil company, published their CSR report over 2009. It prominently featured images showing the firm’s responsible behavior. BP later plead guilty on charges relating to the crash, also admitting it had misled investors in its aftermath (Krauss & Schwartz, 2012).
Only recently researchers have referred to the misalignment between CSR disclosure and CSR performance as CSR decoupling (Tashman, Marano, & Kostova, 2019) or, more specifically, the gap between a firm’s internal and external CSR actions (Hawn & Ioannou, 2016). If disclosure does not correspond to performance (i.e. they are decoupled), this undermines the idea of CSR. Moreover, decoupling allows managers to create a tailored image of the firm that appeals stakeholders by exploiting information asymmetry between these parties. This is detrimental for investors because management essentially misleads them by presenting an incomplete and one‐sided idea of CSR performance (Kim & Lyon, 2015). It has indeed been argued that investors might have difficulty in understanding a firm’s real CSR performance and its hypocritical practices (García-Sánchez, Hussain, Khan, & Martínez-Ferrero, 2020). At the same time, the Volkswagen emission scandal in 2015 has demonstrated that decoupling practices are risky. If uncovered, such ‘empty claims of ethical practice’ could spur public anger and cynicism, severely damaging both reputations and share value (Castillo & Vial, 2017). While this is one example, researchers have pointed out that decoupling negatively affects financial performance (Hawn & Ioannou, 2016; Walker & Wan, 2012).
Therefore, it is important to understand decoupling and the mechanisms that may reduce it. This paper contributes to literature by studying the relationship between external CSR assurance (CSRA) as such a mechanism and the extent to which firms can decouple their disclosure from practice. More specifically, it also contributes to literature by studying whether the choice for assurance provider (‘assurance quality’) affects the extent in which firms
decouple. As such, this paper complements both the limited literature on the various determinants of CSR decoupling and general studies on the effectiveness of CSRA, the latter being highly contested (Casey & Grenier, 2015).
Two issues are central to this paper. First, literature has shown that CSR decoupling can be characterized as an information asymmetry problem (Crilly, Zollo, & Hansen, 2012). Because CSR has become important for firms, stakeholders want to understand a firm’s CSR actions and performance (Sauerwald & Su, 2019). Investors, specifically, demand transparency on business activities because they are financed with their capital. However, because management is separated from ownership, they only get to know what managers report. It is assumed that voluntary disclosure reduces this information asymmetry (Cho, Lee, & Pfeiffer, 2013; Jensen & Meckling, 1976). However, it is management who decides what issues to disclose, and what rather to omit. The voluntary nature of disclosure still allows managers to decouple and create a gap between real CSR performance and their disclosure. Some researchers have denoted this practice as ‘hypocritical behavior’ as firms set up ‘organizational facades’ (Cho, Laine, Roberts, & Rodrigue, 2015). Second, studies have shown that external assurance increases user confidence in CSR reporting. Because it ensures that the information indeed reflects reality, thereby making it more credible, CSRA is assumed to reduce information asymmetry (Casey & Grenier, 2015). Therefore, while firms use decoupling mainly for their legitimization, agency theory helps in explaining how CSRA negatively affects how firms are able to decouple. Since external assurance should limit managers’ reporting freedom, it is a more efficient approach than just voluntary disclosure, which still allows decoupling. Independent assurance, therefore, may act as an external monitoring mechanism that reduces decoupling (Sauerwald & Su, 2019). In addition to this, it could be argued that – if the report is assured – assurance quality is higher for Big 4 accounting firms. Application of established accounting standards, the need of maintaining reputational capital, and size benefits (including those of economies of scale and of ‘in-house’ knowledge present in their multiple disciplines) could place them at an advantage compared with other providers.
Although literature suggests that (1) decoupling is an information asymmetry problem that cannot be solved by voluntary disclosure alone and that (2) CSRA improves the quality of disclosure, a possible relationship between CSRA and CSR decoupling remains unexplored. Therefore, the research question of this paper is: What is the effect of external assurance on the extent to which companies are able to decouple their CSR disclosure from CSR performance?
First, I test for a negative relationship between external assurance and CSR decoupling (H1). Subsequently, I study whether assurance quality has a negative effect on CSR decoupling, in which Big 4 accounting firms are considered the higher quality providers (H2). For these tests, a sample of 30,717 firm-years between 2006 and 2016 is used (total number of firms: 4,407). These are worldwide corporations that issued a CSR report and are assigned scores for both their disclosure and performance in the respective years. CSR decoupling is calculated as the absolute difference of standardized disclosure and performance scores. Furthermore, for a total of 13,555 observations it is indicated whether the report is assured and, if so, the name of the assurance firm is provided.
Descriptive statistics show that understatement of CSR performance is far more common than its overstatement. Results of the regression analyses indicate that CSR decoupling is negatively affected if a CSR report is assured. Moreover, if a report is assured, higher assurance quality is also found to effectively reduce CSR decoupling. The latter finding is, however, not confirmed by the robustness test. Additional analyses provide further insight, as both hypotheses are supported by evidence for the understating subsample and for environmental and social decoupling. However, for the overstating subsample, CSRA is found to have a positive effect, while assurance quality has no effect. Hypotheses are also not supported for governmental decoupling, as no effect is found.
The next section presents a literature review and two hypotheses. In the method section the sample, methodology and measures for the empirical analysis are described. This is followed by the results section in which the findings are presented. These are discussed afterwards. The final section summarizes the paper and suggests future research directions for studying the determinants of CSR decoupling.
2. Theory and Hypothesis Development 2.1 CSR Performance
Over the past decades, initiatives on environmental, social, and governance (ESG) performance have increased significantly, often referred to as CSR actions (Hawn & Ioannou, 2016). The main drivers for engaging in these actions are external pressure, the perspective of economic benefits, and ethical considerations (Graafland & Smid, 2019). As companies recognize both the pressure and advantages, they spend more resources than ever on CSR activities (Du,
Bhattacharya, & Sen, 2010). One important characteristic of these initiatives is that they are employed on a voluntary basis, that is, they are “actions that appear to further some social good, beyond the interests of the firm and that which is required by law” (McWilliams & Siegel, 2001, p. 117). Researchers have pointed out that this can both mean that firms ‘do more good’ and/or that they ‘do less harm’ (Fu, Tang, & Chen, 2019). From a neo-institutional perspective, specifically, CSR performance has been referred to as a firm’s internal actions (Hawn & Ioannou, 2016) or substantive actions (Schons & Steinmeier, 2016). They are initiatives such as policies and guidelines fighting corruption or tax evasion, the formation of committees, and training programs (Hawn & Ioannou, 2016). These actions should result in measurable changes in operations, often at higher costs (Schons & Steinmeier, 2016). Recent studies have assessed this performance by using scores from databases such as KLD (Casey & Grenier, 2015; Dhaliwal, Li, Tsang, & Yang, 2011; García-Sánchez et al., 2020). More than one hundred CSR rating schemes existed in 2014, which compete by aiming to both imitate and differentiate from one another (Pope & Wæraas, 2016). Although scores have been adopted as appropriate measures, literature has noted that assessing CSR performance keeps being inherently subjective (Cho, Lee, & Pfeiffer, 2013).
2.2 CSR Disclosure
Not only engaging in CSR actions is voluntary for companies, they are generally also free in whether and how to report on these activities (Casey & Grenier, 2015). Firms may be required to disclose some nonfinancial information, but no regulation exists for issuing reports (Simnett, Vanstraelen, & Chua, 2009). Companies mainly choose to communicate their CSR initiatives in reports, corporate websites and digital channels such as social media, the latter providing a distinct advantage over the others as it facilitates direct stakeholder interaction (Castillo & Vial, 2017). As efforts have increased, corporations are increasingly open about their social and environmental impact (Chan, Watson, & Woodliff, 2014). Sustainability reports published by the world’s 250 largest corporations went from 35 to 93 percent between 1999 and 2017 (KPMG, 2017). From the neo-institutional perspective, CSR disclosure has been considered part of a firm’s external actions (Hawn & Ioannou, 2016) or symbolic actions (Schons & Steinmeier, 2016). External actions include all forms of communication and visible initiatives (García-Sánchez et al., 2020, p. 5) such as branding, reporting, environmental protection or eliminating child labor (Crilly et al., 2012; Hawn & Ioannou, 2016; Pope & Wæraas, 2016). Contrary to substantive actions, symbolic actions do not require significant investments or
changes at the operational level, while still at least appearing to meet stakeholder demands (Schons & Steinmeier, 2016).
Why would firms bother publishing internal information when this is not required? Several theories are used to explain this decision. Voluntary disclosure theory holds that firms have an internal incentive to signal their proactive strategy to stakeholders if they perform better on CSR measures. This implies that disclosure is indicative of the organization’s CSR commitment (Boiral, 2013). However, most theories recognize the importance of external pressure, which refers to stakeholder demands of (1) transparency and (2) legitimization (Hawn & Ioannou, 2016; Schons & Steinmeier, 2016). Ironically, theories on voluntary disclosure thus search for elements that make CSR reporting less voluntary. First, agency theory offers an explanation for the information asymmetry between management and shareholders that occurs when ownership and control of firms are separated. It is assumed that voluntary disclosure can help in resolving agency conflicts by increasing transparency (Jensen & Meckling, 1976). In this respect, it is important to note that CSR reporting has indeed been found to be a useful mechanism in reducing information asymmetry between managers and stakeholders (Cho et al., 2013; Moroney, Windsor, & Aw, 2012). Second, other popular theories are theories of political economy that describe the relation between firms and societies, particularly signaling theory, legitimacy theory, and stakeholder theory (Casey & Grenier, 2015; Cho, Laine, Roberts, & Rodrigue, 2015). In these theories, disclosure is a way to influence the public image of the company “in such a way that the company is regarded a ‘good corporate citizen’ and its actions justify its continued existence (Hooghiemstra, 2000, p. 57).” If this is the case, firms could have an incentive not only to disclose, but also to adapt their disclosure to stakeholder demands.
2.3 CSR Decoupling
CSR decoupling is the misalignment between CSR performance and CSR reporting (Tashman et al., 2019). Just as financial statements should fairly represent a firm’s financial situation, CSR reports should reflect its ‘social imperatives and consequences’ (Matten & Moon, 2008). Therefore, they can be considered the ‘nonfinancial equivalent’ to the financial statements, sharing the same purpose of meeting the public’s information needs (Simnett et al., 2009). However, literature has pointed out that the relationship between CSR performance and reporting is ‘controversial’ (Casey & Grenier, 2015). Despite increasing reporting and apparent transparency, facilitated by guidelines such as the Global Reporting Initiative (GRI), CSR
disclosure does not always correspond to actual CSR performance (Marano, Tashman, & Kostova, 2017; Schons & Steinmeier, 2016). It means that reports are incomplete, biased, or selective and thus of low quality (García-Sánchez et al., 2020). In this way, decoupling limits a stakeholder’s ability to assess a firm’s CSR performance from its disclosure. By not disclosing all factors, or by changing them, management essentially misleads stakeholders. Crilly et al. (2012), however, note that decoupling does not have to be intentional but can just be a ‘muddling through’ process.
In this study, I will follow Hawn & Ioannou (2016) and understand decoupling specifically as the difference between internal and external actions.1 They find that most firms are positively contributing to society and the environment through their operations without fully reflecting this in their external actions, while a minority shows more external than internal actions. A similar conclusion is drawn by Pope & Wæraas (2016), arguing that ‘CSR-washing is rare’ (understood as overstated or exaggerated performance) because it is highly contingent. It is important to recognize that the gap can go two ways, resulting in either overstated or understated CSR. Therefore, some argue that decoupling occurs in “nearly all global CSR initiatives (Pope & Wæraas, 2016, p. 181).” Consequently, although a successful strategy is dependent on various factors, firms can show a ‘mere talk’ or ‘mere walk’ strategy, each either showing more substantive or more symbolic actions (see Figure 1).
Figure 1. CSR strategy matrix, adapted from Schons & Steinmeier (2016)
1 As noted, this distinction is similar to that between substantial and symbolic actions (e.g. Christmann &
Taylor, 2006; Schons & Steinmeier, 2016; Walker & Wan, 2012) which draws on neo-institutional theory (DiMaggio & Powell, 1983; Meyer & Rowan, 1977).
Why do firms decouple? Since decoupling can be characterized as incompleteness of disclosure, it has been explained as an ‘adaptation of reality’ to stakeholder demands by using social-political and organizational theories, particularly legitimacy theory (Brown & Deegan, 1998; MacLean & Behnam, 2010). After all, if voluntary disclosure is seen as a legitimization project, decoupling could be a next step if companies believe that truthful reporting will not match stakeholder expectations. In the case of overstated performance, legitimacy theory thus argues that a firm “purposefully obfuscates potentially controversial actions (Cho et al., 2015, p. 83).” It could also explain understated performance, as shareholders may oppose substantial performance because it is costly (Kim & Lyon, 2015). This accounts for why firms respond differently to stakeholder pressures and adopt a ‘wide range of approaches’ to engaging in external and internal actions (García-Sánchez et al., 2020; Hawn & Ioannou, 2016). However, in both cases, it is the symbolic rather than the substantive actions that determine the legitimacy gained among stakeholders (Schons & Steinmeier, 2016).
Next to answering the why question, agency theory is useful in explaining how a firm is able to decouple. From this perspective, information asymmetry between management and stakeholders allows for a “lack of transparency and credibility on the firm’s internal actions (Hawn & Ioannou, 2016, p. 2571).” In addition to this, the voluntary and unregulated nature of CSR reporting has been argued to be another factor, further increasing a firm’s ability to decouple (García-Sánchez et al., 2020). According to Cho et al. (2015), firms indeed exploit this lack for legitimacy purposes by engaging in hypocritical ‘organizational facades’. Others have indeed concluded that managers are intentionally ‘faking it’ according to stakeholder interests in the presence of information asymmetry (Crilly et al., 2012).
Table 1 presents a literature review of studies on CSR decoupling and CSRA, although the distinction is arbitrary for some papers. Behnam & MacLean (2011), for example, find that CSR decoupling is, among other things, more likely in the absence of sufficient assurance structures, reducing accountability to stakeholders. Similarly, Sauerwald and Su (2019) find that CSRA (as a control variable) reduces CSR decoupling. While in the major online libraries no articles are found that explicitly study the relationship between these concepts, literature in general sheds light on both the determinants of CSR decoupling and CSRA effectiveness.
Table 1. Literature review
Study Main findings Data
source
Applied theory Region
CSR decoupling
Boiral (2013) Ninety per cent of significant negative events are not reported in CSR reports; images show various ‘simulacra’ disconnected with business impact.
Content analysis
Critical theory International
Bromley & Powell (2012)
Instead of a policy-practice gap, decoupling should rather be understood as a means-ends gap.
Literature review
N/A N/A
Cho, Laine, Roberts, & Rodrigue (2015)
Societal and institutional pressures lead to organizational hypocrisy, resulting in decoupling.
Content analysis Signaling theory; legitimacy theory United States Christmann & Taylor (2006)
Firms select their level of compliance by symbolic and substantive use of certifiable standards.
Survey Legitimacy theory China
Crilly, Zollo, & Hansen (2012)
Decoupling may be not intentional and just a ‘muddling through’ in the face of different institutional pressures.
Interviews; ratings; archival
Institutional theory North America and Europe
Delmas & Burbano (2011)
Drivers of greenwashing can be categorized in (non)market external, organizational, and individual psychological drivers.
Literature review
Institutional theory N/A
García-Sánchez, Hussain, Khan, & Martínez-Ferrero (2020)
CSR decoupling leads to higher analysts’ forecast errors, a greater cost of capital, and reduced access to finance.
Archival Stakeholder theory United States
Graafland & Smid (2019)
Evidence points at four levels of nondivergence rather than CSR decoupling; locating CSR responsibility at board level reduces decoupling.
Archival Institutional theory International (24 countries)
Hawn & Ioannou (2016)
Firms undertake more external than internal CSR actions, and a wider gap between the two types is negatively associated with market value.
Archival Neo-institutional theory
International (33 countries)
Jamali, Lund-Thomsen, & Kara (2017)
Coupling and decoupling are dynamic processes which can provide credit and traction for developing country firms.
Surveys; interviews; site visits Institutional theory; critical theory India
Kim & Lyon (2015) Deregulation significantly affects the choice between greenwashing and brownwashing.
Archival Organizational theory
United States
MacLean & Behnam (2010)
Decoupling creates a ‘legitimacy façade’ that enables the institutionalization of misconduct and triggers a loss of external legitimacy.
Literature review Neo-institutional theory; legitimacy theory N/A
Marquis & Qian (2014)
Government dependency leads firms to issue CSR reports; firms are more likely to enact substantive CSR when they are monitored.
Archival Institutional theory China
Marquis, Toffel, & Zhou (2016)
More environmentally damaging firms are less likely to engage in selective disclosure.
Archival Institutional theory International
Pope & Wæraas (2016)
Successful CSR decoupling is rare because many conditions are highly contingent.
Literature review
N/A N/A
Sauerwald & Su (2019)
CSR decoupling is positively related to CEO overconfidence, but this can be mitigated by the board of directors.
Content analysis; archival
N/A United States
Schons & Steinmeier (2016)
The impact of decoupling on firm performance depends on stakeholder proximity.
Archival Neo-institutional theory; stakeholder theory International Tashman, Marano, & Kostova (2019)
Internationalization creates a ‘dual embeddedness’ that drives CSR decoupling.
Archival Neo-institutional International
Walker & Wan (2012)
Substantive environmental actions have no financial consequences, but symbolic actions are negatively related to financial performance.
Archival Legitimacy theory; prospect theory
First, research on CSR decoupling has primarily focused on its determinants, for example activist audit threat (Lyon & Maxwell, 2011), CEO overconfidence (Sauerwald & Su, 2019), firm size and organizational cost (Wickert et al., 2016), and governmental settings (Marquis & Qian, 2014). Findings are often consistent with the two main theories, showing that firms are generally more likely to report good news, and that weaker-performing firms are more likely to report their positive CSR efforts (Casey & Grenier, 2015). Others have examined its consequences on firm value (Hawn & Ioannou, 2016), performance (Schons & Steinmeier, 2016), and other financial market outcomes (García-Sánchez et al., 2020).
Next, although insight in its effectiveness is still limited (Cohen & Simnett, 2015; Simnett et al., 2009), most studies find CSRA to enhance the credibility of reports (e.g. Moroney et al., 2012; Pflugrath et al., 2011). However, other studies are critical and find CSRA itself to be a ‘decoupled or symbolic image of accountability’ that undermines its credibility as a verification practice (Perego & Kolk, 2012). In its absurd but disturbing extreme, this could leave both the reporting and assurance mechanism to be a “hyperreality conveying signs, data and images without any reference to the real world (Boiral, 2013, p. 1043).” The limited literature on CSRA
Weaver, Treviño, & Cochran (1999)
External pressures encourage decoupled processes, but management can encourage both easily decoupled and integrated processes.
Survey; archival
Institutional theory United States
Wickert, Scherer, & Spence (2016)
Organizational costs for CSR talk decrease with firm size, providing financial incentives to focus on this aspect rather than on costly CSR walk.
Literature review Theory of the firm; contingency theory N/A
Winkler, Etter & Castelló (2020)
Self-persuasive instead of agonistic rhetoric drives CSR decoupling.
Literature review
N/A N/A
CSR assurance
Casey & Grenier (2015)
CSRA is associated with lower cost of equity capital and lower analyst forecast errors and dispersion; these effects are significantly higher when CSRA is provided by an accounting firm.
Archival Meta-theoretical perspective United States/ International Moroney, Windsor, & Aw (2012)
CSR disclosure quality is higher for assured companies; CSRA quality does not differ between accounting and consultant providers.
Archival Stakeholder-agency theory
Australia
O’Dwyer & Owen (2005)
CSRA independence is questionable as there is a large degree of management control over the assurance process.
Archival N/A UK & Europe
Perego & Kolk (2012)
Firms project a decoupled or symbolic image of accountability through CSRA.
Archival Neo-institutional theory
International
Pflugrath, Roebuck, & Simnett (2011)
CSR disclosure credibility is greater when it is assured and when the assurer is an accountant.
Behavioral experiment
N/A Australia, UK, United States, Simnett,
Vanstraelen, & Chua (2009)
The decision whether to seek CRSA is more important for firms than whether the CSRA provider is a member of the auditing profession.
Archival Stakeholder theory International
Zorio, García-Benau, & Sierra (2013)
CSRA quality is significantly higher if the provider is an auditor and with larger firms.
has been inconclusive on the point of quality differences between providers. While there is substantial variability in scope, independence, criteria, and assurance levels, some researchers have not found a significant difference (Moroney et al., 2012). Others concluded that accounting firms indeed outperform other providers (Pflugrath et al., 2011; Zorio et al., 2013). Perego & Kolk (2012) find that superiority varies between aspects such as format, procedures, and recommendations. While these studies have stressed the importance to ‘mind the gap’ and have provided useful insights in determinants of assurance quality, still very little has been researched about the role of external audit in reducing CSR decoupling specifically.
2.4 Hypothesis Development
Reducing decoupling, whether intentional or emergent, is important for investors because they provide managers the capital necessary to undertake substantive CSR actions. Prior studies have shown that investors seek reliable information about CSR performance to base their decisions upon (Cohen, Holder-Webb, Nath, & Wood, 2011; Cohen & Simnett, 2015). If managers mainly choose to engage in symbolic actions, they are essentially stealing money from investors. This is true for both directions of the gap. Indeed, studies show that managers are punished for their hypocritical practices. As Hawn & Ioannou (2016) argue, decoupling negatively affects market value because overstating CSR increases the risk of being exposed, while understating CSR limits an investor’s ability to assess the value-creating potential of substantive actions. Furthermore, García-Sánchez et al. (2020) show that decoupling destroys shareholder value by increasing cost of capital and reducing a firm’s access to finance. Investors, therefore, want to be able to assess a manager’s performance regarding CSR, and not only in terms of returns.
As noted above, agency theory provides a useful explanation of how a firm’s ability to decouple arises. From this perspective, the presence of information asymmetry between management and stakeholders is a condition for decoupling. Indeed, Pope & Wæraas (2016, p. 177) show through their CSR-washing diagram that for firms to successfully decouple, it is necessary that users are not able to distinguish true and false CSR advertisements. Consequently, decoupling has been characterized as an executive strategy in which management takes advantage of information asymmetry (Crilly et al., 2012). If this is the case, agency theory will also be helpful in explaining how of a firm’s ability to decouple can be limited. In other words, when thinking of mechanisms that could reduce decoupling, it seems logical to look for means that reduce information asymmetry. If firms are able to decouple, CSR talk is often perceived as
‘cheap talk’ and cynically received, leading to a ‘circle of disengagement’ (Hawn & Ioannou, 2016; Winkler et al., 2020). As Manetti & Becatti (2009, p. 289) put it, “there is a credibility gap, that renders sustainability reports an instrument little used by the traditional target users.” For this reason, voluntary disclosure alone is not enough in solving the problem of information asymmetry – and could even increase legitimacy issues.
In this study, therefore, I follow Manetti & Becatti (2009) in turning to the ‘strategic role’ of assurance as an instrument to address CSR decoupling (cf. Dando & Swift, 2003). Because decoupling is misleading and destroys value, investors are concerned about decoupling and increasingly demand assurance of CSR reports, which ‘breaks down the hegemonic discourse’ (Edgley, Jones, & Solomon, 2010; Moroney et al., 2012). Because managers will eventually be punished by investors if it turns out that disclosure does not correspond to performance, they should take the auditor’s opinion seriously.
External Assurance and CSR Decoupling
Not only (1) engagement and (2) reporting of CSR activities are voluntary, firms are also completely free to have their reports (3) assured or not (Manetti & Becatti, 2009; Perego & Kolk, 2012). An important source of information on external assurance of CSR reports is the KPMG survey on sustainability reporting. Besides showing a growing number of companies reporting worldwide, their 2017 report also indicates that the proportion of firms investing in some form of third-party CSR assurance has grown steadily to 67 percent (KPMG, 2017). The main objective of assurance is “to ensure that the disclosed information is reliable and in compliance with standards, and that it therefore reflects reality as transparently as possible (Boiral, 2013, p. 1039).” CSR assurance is, however, complicated by the fact that the diversity of subject matter, lack of analytic tools, and a lack of criteria could cause ‘substantial difficulty’ for CSRA providers in assessing reports, even though there is similarity in the risk-based approach. This extends to risk identification, materiality, ability to identify misstatements and fraud, evidence collection, assurance reports and communication (Casey & Grenier, 2015). However, studies have found that quality and credibility of CSR reports have been improved by standards that have emerged in the last decade (Perego & Kolk, 2012), especially the GRI Sustainability Reporting Framework, ISAE 3000, and AA 1000 AS, the latter mostly used by non-accounting firms (Manetti & Becatti, 2009).
The central argument of this paper, considering information asymmetry as a condition for decoupling, is that CSRA is an effective response to decoupling practices because it enhances the disclosed information’s accuracy, validity, completeness, and reliability (Boiral, 2013; KPMG, 2017; Simnett et al., 2009). In this way, CSRA increases transparency by reducing information asymmetry and agency costs, making communications more effective (Casey & Grenier, 2015; Perego & Kolk, 2012). Consequently, assurance functions as an external monitoring mechanism that reduces mere symbolic CSR expression (Sauerwald & Su, 2019). This makes the information more value relevant (García-Sánchez et al., 2020). Indeed, some studies find that CSR reporting quality is significantly higher for assured companies than for unassured companies (Moroney et al., 2012; Pflugrath et al., 2011). Therefore, I argue that independent assurance reduces CSR decoupling and hypothesize the following:
H1: There is a negative association between external CSR assurance and CSR decoupling, ceteris paribus.
Assurance Quality and CSR Decoupling
Next to the voluntary engagement, disclosure and assurance, firms are often also free to choose their assurance provider. This has resulted in a largely unregulated and competitive market – unlike that of financial statement assurance – in which both accounting firms and non-firms have significant market shares (Cohen & Simnett, 2015), the latter type including environmental consulting firms, NGOs and certification bodies (O’Dwyer, Owen, & Unerman, 2011). Literature shows that approaches and expertise between the two types of assurers vary considerably (O’Dwyer & Owen, 2005; Perego & Kolk, 2012). Specifically, accounting firms prefer a financial statement audit style with particular interest in the information’s material accuracy, while nonaccounting firms offer more qualitative audits (Casey & Grenier, 2015). This competition and variability provide opportunities for research into quality differences. Several criteria determine the eventual quality of an assured report. These are independence of the assurer, sufficient understanding and expertise, and the implementation of quality controls (Huggins, Green, & Simnett, 2011). As Cohen & Simnett (2015, p. 61) note, it “is only in these circumstances that CSR activities will be substantively effective and not mere public relations exercises.” Building on these aspects, researchers have pushed forward several arguments why Big 4 firms can be considered the higher quality providers and are thus ‘well placed’ for delivering CSRA services (Huggins et al., 2011), even though they may not always fully
develop their skills (Gray, 2000). A first argument is that the auditing profession already has well-developed standards, for example on engagement acceptance, required expertise, and use of experts. This has resulted in a comprehensive body of ethics, independence and objectivity requirements, and quality control mechanisms (Simnett et al., 2009). For instance, the ISAE 3000 standard parallels the financial statement audit risk model and stresses the need for competencies in both assurance and the related subject matter (Cohen & Simnett, 2015; Huggins et al., 2011). In addition, when knowledge is not present, external experts can be bought or hired (Simnett et al., 2009), and it has been argued that acquiring subject matter expertise will be easier for accounting firms than it is for non-firms to acquire assurance expertise (Pflugrath et al., 2011). Second, because they have an established history and reputational capital, Big 4 firms will put relatively more effort to maintain this capital and hence behave less opportunistically, which increases confidence in their competency and legitimacy as high-quality assurance providers (Simnett et al., 2009). Although this often means that Big 4 assurance comes at a higher cost, price could be an ‘indication of quality’ if the market is efficient (Simnett et al., 2009). A third category of arguments stresses the size of Big 4 firms. One argument is that Big 4 firms will be less affected by fee dependence because of their size. In addition to this, with regard to specialization in general, their multidisciplinary teams will be easier to compose because of their global reach, and more streamlined when engagements require to work internationally (Carson, 2009). Also related to size is the argument that Big 4 firms are better able to invest in new technologies because of their scale economies (Simnett et al., 2009). The latter is particularly relevant for assurance of CSR reports, as firms are faced with a ‘new’ variety of subject matter which requires different analytic tools. Accordingly, it can be expected that these aspects work to their advantage in detecting and tackling decoupling practices.
Considering decoupling as issuing low-quality reports (Marquis & Qian, 2014), Big 4 firms can be expected to deliver higher assurance quality, leading to more reliable reports and less decoupling. Hence, following Simnett et al. (2009), I classify Big 4 accounting firms as the higher quality CSRA providers. Therefore, I hypothesize the following:
H2: There is negative association between assurance quality and CSR decoupling, ceteris paribus.
3. Methodology 4.1 Sample and Data Sources
This study focuses on publicly listed corporations worldwide that have published a CSR report, in the first place because both firm characteristics and score data are available for these firms. This is in line with prior archival research on CSR decoupling. Furthermore, as argued, the importance of CSRA mainly stems from its relevance for investors in their investment decisions. Next, an international instead of a regional focus is desirable because it increases external validity. It is also justifiable in the context of this paper because both engagement in internal and external CSR actions and assurance on disclosure are worldwide phenomena (Hawn & Ioannou, 2016; KPMG, 2017).
Archival data is obtained from the Thomson Reuters ASSET4 ESG and Bloomberg databases, with additional control variables downloaded from the WorldScope database. The initial merged dataset contains 97,800 observations. Because most observations are too fragmented before 2006, the window for this study is set from 2006 until 2016. By using cross-sectional and time series (panel) data, based on firm and year, chances of collinearity among the variables are reduced. Missing observations for the control variables were replaced by the industry mean, except for board-level variables which were replaced by the firm mean (mode for categorical variables). After this, observations missing any of the dependent and control variables were removed. Next, to reduce the effect of outliers, continuous variables are winsorized at the 1% and 99% level. After these adjustments, a final sample of 30,717 observations remains. Not all firms are represented in every year. The sample consists of 4,407 unbalanced panels (firms) with a maximum of 11 firm-years each (2006-2016), including firms from 58 countries and all industries if classified according to the SIC-based Fama-French industry classification. Details on industry and geographical representation are included in Appendix A.
4.2 Variables and Measures Dependent variable
CSR decoupling is measured as the absolute value of the CSR disclosure score (t) minus the CSR performance score (t-1). Because a current gap is determined by a firm’s previous internal actions, lagged values of performance scores are used. Following Tashman et al. (2019), scores are standardized before calculating the gap. Because standardized scores are subtracted, the
obtained gap variable is also measured in standard deviations. Next, in line with Hawn & Ioannou (2016), this variable is converted to absolute values because this allows for capturing the misalignment in both directions.
CSR performance scores are obtained from the Thomson Reuters ASSET4 ESG database, which provides z-scores (ranging from 0 to 100) that sets off a firm’s performance against other firms (Hawn & Ioannou, 2016). Scores are based not only on CSR reports, but also on information found in general annual reports, websites, proxy filings, and news sources (Schons & Steinmeier, 2016). The overall score used is based on the individual ESG scores, comprised of ten evaluation points, plus a so-called ‘ESG controversy’ score (Thomson Reuters, 2020). Although ASSET4 also offers disclosure data, the gap variable for the main analysis is calculated using disclosure data from the Bloomberg ESG database. Bloomberg rates firms on a similar scale but evaluates the data with respect to industry, which makes it more informative (García-Sánchez et al., 2020). Next to these overall scores, both databases also provide the underlying scores for all three ESG categories. For the additional analyses, a similar method is followed in calculating decoupling for each category.
Independent variables
Data on the independent variables is also obtained from ASSET4. Both ‘External assurance’ and ‘Big 4 assurance’ are indicator variables. Of all observations, the choice whether to obtain CSR assurance is indicated for 13,555 firms. For all firm-years with external assurance, the name of the provider is listed. In order to operationalize these variables, ‘External assurance’ will be assigned a value of 1 if the CSR report is assured, and 0 otherwise. Similarly, ‘Big 4 assurance’ will be assigned a value of 1 if firms have their CSR report assured by a Big 4 firm, and 0 otherwise. As ‘Big 4’ were classified all providers that contained the name (or any abbreviation) of Ernst and Young, PricewaterhouseCoopers, KPMG, and Deloitte & Touche. Logically, ‘Big 4 assurance’ only has values if firms have their CSR reports assured. This is consistent with other studies on or including the effect of CSRA (e.g. Moroney et al., 2012; Sauerwald & Su, 2019; Simnett et al., 2009).
Control variables
Profitability, size, and other firm-specific characteristics have been argued to influence a firm’s available decoupling strategies (Delmas & Burbano, 2011). I will therefore add several variables to control for their assumed effect on CSR decoupling. Profitability will be controlled
for by including return on assets (ROA), because better performing firms will have easier access to capital and will be less reluctant to invest (McWilliams & Siegel, 2001; Peters & Romi, 2014). ROA is calculated as net income divided by total assets. Next, I add firm size as the log of total assets, as larger firms are better able to attract resources for investments (Peters & Romi, 2014) and because their social and environmental impact is more closely followed (Christmann & Taylor, 2006). Sales growth, as the percentage of change in sales with respect to the previous year, accounts for a firm’s capability to invest in substantive CSR actions. Capital intensity, as the ratio of assets to sales, can be indicative of the way in which firms use their assets to improve CSR performance (Tashman et al., 2019). Next, I will include R&D intensity because the amount of research and development (R&D) can stimulate substantive CSR actions (McWilliams & Siegel, 2001). It is measured as R&D expenses divided by net sales. Next, financial leverage, measured as total liabilities divided by total assets, is included because higher levels of debt imply that firms have fewer resources for substantive CSR actions (Marano et al., 2017). Moreover, higher debt levels also lead to increased monitoring and thus create an incentive to disclose information (Peters & Romi, 2014).
Next to these, four board-level variables are included to control for corporate governance effects. Rationales for these controls are suggested by two other studies on, respectively, CSR decoupling (Sauerwald & Su, 2019) and sustainability performance as the product of a disclosure index and a quality index (Hussain, Rigoni, & Orij, 2018). Board size, measured as the total number of directors, is added because larger boards are considered less effective in monitoring governance, particularly because of higher coordination costs and less involvement from individual directors. Board independence, as the percentage of outside directors on the board, is used because outside directors could positively impact CSR performance by critically monitoring the CEO and because they represent other stakeholders. Board gender diversity, measured as the percentage of women on the board, is included because literature indicates that women on boards, as compared with men, are more socially responsive and less inclined to show symbolic actions. CEO duality, a binary variable that has value 1 if the CEO is also the chairman of the board 0 otherwise, is included because the influence of a CEO is greater when these functions are combined, which may influence reporting decisions.
Table 2. Variables, measurement, and data sources
4.3 Research Model
The first hypothesis tests the effect of external assurance of CSR reports on CSR decoupling. The following model is used to test Hypothesis 1:
CSRDit = 𝛼 + β1CSRAit + β2ROAit + β3SIZEit + β4SGRTHit+ β5CAPINTit+ β6RDINTit+ β7LEVit+ β8BSIZEit+ β9BINDPit+ β10BGENit+ β11CEODit+ εit
If firms choose for external assurance, the second hypothesis tests whether the extent of decoupling differs between Big 4 and other assurance providers. The model used to test Hypothesis 2 is equal to that of Hypothesis 1, except that CSRA is replaced by BIG4 as the independent variable.
In these models, βit refers to the variable coefficients and εit is the combined error term. The presence of i and t indicates the individual and time dimensions of panel data. Using panel data
Variable Acronym Measurement Source
Dependent variable
CSR Decoupling CSRD Absolute value of CSR disclosure score (t) minus CSR performance score (t-1)
ASSET4/ Bloomberg
Independent variables
External Assurance CSRA Binary variable that equals 1 if the CSR report is independently audited, and 0 otherwise
ASSET4 Big 4 Assurance BIG4 Binary variable that equals 1 if the CSR report is
audited by a Big 4 firm, and 0 otherwise
ASSET4
Control variables
Profitability ROA Ratio of net income and total assets WorldScope Firm Size SIZE Log of total assets WorldScope Sales Growth SGRTH Percentage of change in sales with respect to t-1 WorldScope Capital Intensity CAPINT Ratio of total assets and net sales WorldScope R&D Intensity RDINT Ratio of R&D expenses and net sales ASSET4 Financial Leverage LEV Ratio of total debt and total assets ASSET4 Board Size BSIZE Number of directors on the board ASSET4 Board Independence BINDP Percentage of outside directors in relation to the
total number of directors
ASSET4 Board Gender Diversity BGEN Percentage of women on the board in relation to
the total number of directors
ASSET4 CEO Duality CEOD Binary variable that equals 1 if the CEO is also
the chairperson of the board, and 0 otherwise
leads to two error terms: an idiosyncratic (varying within and between individuals, as in unpaneled data) and an individual-specific error term (constant within each individual but varying between). The latter is especially important, because its assumptions determine whether a fixed-effects or random-effects regression model is appropriate. While a pooled OLS approach does not take this issue into account, in the fixed (random) model, the individual-specific effect is correlated (uncorrelated) with the explanatory variables. These assumptions can be tested by performing a Hausman test. Untabulated results show that this correlation is not zero for both models (p < .000), indicating that a fixed-effects model will better fit the data. A major disadvantage of this decision is, however, that time-invariant effects (e.g. that of country and industry) cannot be estimated separately. Next, because a fixed-effect model is used, a modified Wald test is performed to test for groupwise heteroskedasticity. Untabulated results show that variances across individuals are not zero for both models (p < .000), so the robust estimator of variance will be used to correct for heteroskedasticity.
4. Results
This chapter presents the statistical results. An overview of the descriptive statistics will be followed by the results of the fixed-effects panel regression models employed to test the hypotheses. Additional analyses are performed by testing on subsamples of each gap direction and by tests for each ESG category. Certainty of the models is assessed by a robustness check using the decoupling data of the study by Hawn & Ioannou (2016), which uses ASSET4 scores for both disclosure and performance. Stata is used to analyze the data.
Descriptive statistics are summarized in Table 3 and Table 4. Table 3 is included first because it provides insight in the relative differences on which the dependent (absolute) variable CSRD is based. CSR overstatement means that the disclosure score is higher than the performance score (difference > 0). Similarly, CSR understatement means that the performance score is higher than the disclosure score (difference < 0). These statistics evidently show a negative CSR gap. Only 20.1% of observations can be classified as overstating, while 79.9% apparently has understated CSR performance. For the difference calculated with unstandardized scores, an untabulated t-test additionally ensures that the sample mean is different from zero (p < .000). Appendix B further illustrates the underlying distribution.
Panel A in Table 4 shows the descriptive statistics of the binary variables. For the dependent variables, assurance levels and the shares of both provider types are presented for each year.
Of all observations, 58.4% of CSR reports has obtained external assurance. Of this percentage, 42.8% is provided by a Big 4 assurance firm. As should follow from the measurement, the frequency of Assured is in all instances equal to the frequencies of Big 4 and Non-Big 4 combined. Assurance levels have risen from 13.8% in 2006 to 81.1% in 2016. In the same period, the proportion of Big 4 accounting firms in CSRA increased from 32.1% to 46.7%. The number of observations per year is stable and ranges between 1,064 and 1,580. Appendix C shows an overview of assurance adoption for overstating and understating firms, respectively. Panel B in Table 4 presents the descriptive statistics of the continuous variables. CSRD, as the absolute value of the gap variable calculated with standardized scores, has an average value of 0.66. Average return on assets is 4%, while average sales growth is 7%. Underlying the SIZE variable is total assets in US$ with a mean of 43.6 million US$ and a standard deviation of 1.87 billion US$. As a measure of capital intensity, net sales on average equals to 4.65 times the book value of total assets. R&D intensity has an average of only 1% with a maximum of 39%. Leverage ratios are as high as 0.80, with firms financing an average of 26% of total assets with debt. The number of board members ranges between 4 and 25, with an average of 11 members. Of these members is, on average, 72.35% independent and 11.95% (and never more than 50%) female. As the only binary control variable, CEO duality is not listed separately but statistics show that in 40.8% of all observations the CEO is also the chair of the board (N = 30,711). Table 5 shows the Pearson correlation coefficients for the variables of Hypothesis 1. Because CSRA and BIG4 are perfectly collinear as a consequence of their measurement, one will be dropped from the correlation matrix when using listwise deletion. Therefore, correlations cannot be presented in one table. Results demonstrate that many variables are significantly correlated, which reflects the firm dynamics in which the relationships between both firm-level and board-level variables are complex. Correlation coefficients higher than 0.800 are considered to indicate multicollinearity (Hawn & Ioannou, 2016). None of the variables of the models for both Hypothesis 1 and Hypothesis 2 (untabulated) show correlations that exceed this limit, so multicollinearity is not assumed.2 This eliminates the need for any additional correlation analyses.
2 Following Moroney et al. (2012), who deploy similar independent variables in their respective hypotheses
Table 3. Relative differences
Difference is CSR disclosure score (t) minus CSR performance score (t-1). See also Appendix B.
a Calculated with standardized scores. Absolute values are used as the dependent variable CSRD in analyses. b Calculated with unstandardized scores. This variable is used for distinguishing understating and overstating firms.
Table 4. Descriptive statistics Panel A: Binary variables
CSRA BIG4A
Assured Non-assured Big 4 Non-Big 4
Year N Frequency % Frequency % Frequency % Frequency % 2006 1,580 218 13.8 1,362 86.2 70 32.1 148 67.9 2007 1,063 309 29.0 754 70.9 112 36.3 197 63.8 2008 1,106 387 35.0 719 65.0 154 39.8 233 60.2 2009 1,092 483 44.2 609 55.8 196 40.6 287 59.4 2010 1,258 592 47.1 666 52.9 248 41.9 344 58.1 2011 1,211 760 62.8 451 37.2 333 43.8 427 56.2 2012 1,086 891 82.0 195 18.0 396 44.4 495 55.6 2013 1,188 989 83.3 199 16.8 428 43.3 561 56.7 2014 1,248 1,042 83.5 206 16.5 437 41.9 605 58.1 2015 1,307 1,088 83.2 219 16.8 472 43.4 616 56.6 2016 1,416 1,149 81.1 267 18.9 536 46.7 613 53.4 Total 13,555 7,908 58.3 5,640 41.7 3,382 42.8 4,526 57.2
Panel B: Continuous variables Mean SD Min. Max.
CSRD 0.66 0.55 0.00 4.53 ROA 0.04 0.08 -0.46 0.30 SIZE 15.83 1.64 11.65 21.00 SGRTH 0.07 0.22 -0.56 1.33 CAPINT 4.65 8.22 0.29 61.43 RDINT 0.01 0.04 0.00 0.39 LEV 0.26 0.13 0.00 0.80 BSIZE 10.58 3.52 4.00 25.00 BINDP 72.35 24.81 0.00 100.00 BGEN 11.95 11.45 0.00 50.00
CSRD is the absolute value of the difference calculated with standardized scores. See also Appendix B.
N = 30,717
N Mean SD Min. Max. Differencea 30,717 0.00 0.86 -3.03 4.53
Differenceb 30,717 -23.83 24.51 -92.38 61.98
Difference > 0 (Overstatement) 6,180 10.69 9.47 0.00 61.98 Difference < 0 (Understatement) 24,537 -32.52 18.81 -92.38 0.00
Table 5. Pearson correlation Variable 1 2 3 4 5 6 7 9 10 11 12 13 1. CSRD 1 2. CSRA -0.046 *** 1 3. ROA -0.014 -0.092 *** 1 4. SIZE -0.003 0.234 *** -0.236 *** 1 5. SGRTH 0.003 -0.098 *** 0.203 *** -0.062 *** 1 6. CAPINT -0.026 ** 0.046 *** -0.195 *** 0.437 *** -0.042 *** 1 7. RDINT 0.008 -0.009 0.026 ** -0.124 *** 0.019 * -0.063 *** 1 8. LEV 0.002 0.039 *** -0.149 *** 0.039 *** -0.027 ** -0.011 -0.012 1 9. BSIZE 0.015 0.127 *** -0.113 *** 0.425 *** -0.030 *** 0.128 *** -0.030 *** 0.029 *** 1 10. BINDP -0.003 -0.019 * 0.060 *** 0.137 *** 0.002 0.086 *** 0.068 *** 0.053 *** 0.052 *** 1 11. BGEN -0.065 *** 0.102 *** 0.020 * 0.132 *** -0.083 *** 0.083 *** 0.040 *** 0.014 0.061 *** 0.462 *** 1 12. CEOD -0.009 -0.105 *** 0.011 0.083 *** -0.016 -0.080 *** -0.089 *** -0.000 0.069 *** -0.106 *** -0.050 *** 1 N = 13,555
Table 6 presents the fixed-effects regression models. All models show only slight explanatory power, with R2 not exceeding 0.027. Because of the collinearity of CSRA and BIG4, the effect
of both variables cannot be estimated simultaneously. Therefore, such a model is not included. Model (1) shows the effect of only the control variables on CSR decoupling. Significant firm-level variables are capital intensity (negative, p < .01) and R&D intensity (positive, p < .05). Significant board-level variables are board size (positive, p < .05) and board gender diversity (negative, p < .001). Next, Model (2) includes the effect of external assurance on CSR decoupling. The coefficient for CSRA is negative (-0.154) and significant (p < .001). This means that external assurance reduces CSR decoupling. Therefore, H1 is supported. With respect to Model (1), the effect of control variables has somewhat changed. The significance level of R&D intensity has increased (p < .001), while board size shows no significant effect when external assurance is included. In Model (3) the independent variable CSRA is replaced by BIG4, indicating whether the assurance provider is a Big 4 accounting firm. Results show that the coefficient for BIG4 is negative (-0.060) and significant (p < .01), indicating that Big 4 assurance specifically reduces decoupling. Hence, H2 is also supported. In this model, effects of R&D intensity and board gender diversity are equal to those estimated in Model (2). Furthermore, capital intensity has no significant effect when the effect of Big 4 assurance is included, while board independence has a negative and significant (p < .05) effect.
Table 6. Fixed-effects regression models
Variable Exp. Sign (1) (2) (3) CSRA ( – ) -0.154 *** (0.02) BIG4 ( – ) -0.060 ** (0.02) ROA ( – ) 0.058 (0.05) 0.088 (0.11) -0.018 (0.15) SIZE ( – ) -0.005 (0.01) -0.024 (0.02) -0.040 (0.03) SGRTH ( – ) 0.003 (0.01) 0.037 (0.03) 0.026 (0.04) CAPINT ( – ) -0.003 ** (0.00) -0.009 ** (0.00) -0.008 (0.00) RDINT ( – ) 0.399 * (0.13) 0.740 *** (0.20) 0.761 *** (0.21) LEV ( – ) -0.006 (0.03) -0.087 (0.06) -0.030 (0.07) BSIZE ( + ) 0.005 * (0.00) 0.004 (0.00) 0.007 (0.00) BINDP ( – ) 0.000 (0.00) -0.001 (0.00) -0.002 * (0.00) BGEN ( – ) -0.002 *** (0.00) -0.004 *** (0.00) -0.004 *** (0.00) CEOD ( + ) -0.012 (0.01) -0.014 (0.02) 0.015 (0.02) Constant 0.711 *** (0.20) 1.276 *** (0.37) 1.494 ** (0.58) R2 0.003 0.027 0.014 Observations 30,717 13,555 7,908 Groups 4,407 2,530 1,400
Table 7 shows the fixed-effect regressions for both a CSR overstatement and CSR understatement subsample, as summarized in Table 3. Results for Model (2) and Model (5) demonstrate that the coefficient of CSRA is positive and significant (p < .001) if firms overstate, and negative and significant (p < .001) if firms understate their CSR performance. These estimations show that adoption of external assurance is related to an increase of decoupling in overstating firms, while related to a decrease in understating firms. For overstating firms, the effect of firm size is positive and significant (p < .001). Furthermore, for understating firms, a negative and significant effect is found for both capital intensity (p < .01) and board gender diversity (p < .001), while R&D intensity shows a positive and significant (p < .001) effect. Table 7. Fixed-effects regression models with subsamples
Dependent variable: Overstatement (rel. CSRD > 0) Understatement (rel. CSRD < 0)
Variable (1) (2) (3) (4) (5) (6) CSRA 0.438 (0.08) *** -0.158 (0.02) *** BIG4 -0.032 (0.08) -0.050 (0.02) * ROA -0.254 (0.11) * -0.600 (0.34) -0.690 (0.47) 0.074 (0.06) 0.065 (0.11) -0.068 (0.15) SIZE 0.139 (0.03) *** 0.258 (0.08) *** 0.308 (0.19) -0.007 (0.01) -0.041 (0.03) -0.028 (0.04) SGRTH -0.049 (0.02) * -0.032 (0.07) 0.353 (0.15) * -0.003 (0.02) 0.027 (0.03) 0.004 (0.04) CAPINT -0.002 (0.00) -0.001 (0.00) 0.014 (0.01) -0.004 (0.00) ** -0.008 (0.00) ** -0.008 (0.00) RDINT 0.292 (0.38) 0.543 (0.68) -0.053 (0.41) 0.479 (0.14) ** 0.753 (0.21) *** 0.809 (0.21) *** LEV 0.056 (0.07) 0.016 (0.19) -0.002 (0.25) -0.027 (0.04) -0.079 (0.06) -0.024 (0.07) BSIZE -0.007 (0.01) -0.005 (0.01) 0.049 (0.02) * 0.006 (0.00) * 0.005 (0.00) 0.005 (0.00) BINDP 0.001 (0.00) -0.001 (0.00) -0.008 (0.00) -0.000 (0.00) -0.000 (0.00) -0.001 (0.00) BGEN 0.007 (0.00) *** 0.003 (0.00) -0.003 (0.01) -0.002 (0.00) ** -0.004 (0.00) *** -0.003 (0.00) ** CEOD -0.062 (0.03) * -0.062 (0.08) -0.126 (0.17) -0.011 (0.01) -0.023 (0.02) 0.000 (0.02) Constant -1.200 (0.44) ** -2.837 (1.19) * -3.183 (3.07) 0.688 (0.23) ** 1.485 (0.41) *** 1.218 (0.59) * R2 0.025 0.140 0.116 0.003 0.029 0.011 Observations 6,180 1,334 415 24,537 12,221 7,493 Groups 2,199 776 263 3,491 2,228 1,325
Results for Model (3) and (6) provide no evidence for a relationship of Big 4 assurance with CSR decoupling if firms overstate their performance. However, for understating firms, the coefficient of BIG4 is again negative and significant (p < .05). This implies that Big 4 assurance specifically is related to a decrease of CSR decoupling only in understating firms, which constitute the majority of the sample. Results additionally indicate effects for several control variables. If firms overstate, a positive and significant effect is estimated for both sales growth (p < .05) and board size (p < .05). For the understating subsample, again a positive effect is found for R&D intensity (p < .001) and a negative effect for board gender diversity (p < .01). Although signs and statistical significance are important in evaluating a model, coefficients should be assessed against the dependent variable’s scale to get a full understanding of the estimated effects. In these models, both independent variables are binary. This means that the coefficient of CSRA and BIG4 shows the estimated difference between the “0” and “1” coded groups. In the main analysis, the coefficient for external assurance is -0.154. In the additional analysis, coefficients are, respectively, 0.438 for overstating and -0.158 for understating firms. Next, the coefficient for Big 4 assurance is -0.060 in the main analysis. In the additional analysis for understating firms, the effect is comparable and shows a coefficient of -0.050. Since the decoupling variable has a mean of 0.66, correlations correspond fairly to scale of the dependent variable (for the full distribution, refer to Appendix B).
Table 8 presents the results of a second additional analysis, in which the dependent variable is replaced by, respectively, environmental, social, and governmental decoupling. This could provide further insight because the estimated effects could be different for each type of decoupling. For the environmental and social category, both external assurance and Big 4 assurance show a negative relationship with the respective decoupling variable (p < .001 for all four estimations). Therefore, these results correspond to those of the main analysis. For governmental decoupling, however, no significant effect is found for either external assurance or Big 4 assurance specifically. This implies that the negative relationships found in the main analysis do not extend to decoupling on a government level. Of the control variables, sales growth shows a negative effect on both social and governmental decoupling if CSRA is included. R&D intensity is found to have a positive relationship with CSR decoupling in all models. Next, board size has a positive and significant effect on environmental and social decoupling, but a negative effect on governmental decoupling, which is significant for Model (8). The effect of board gender diversity is negative and significant for all models.
Table 8. Additional analysis per CSR category
Dependent variable: ENVD SOCD GOVD
Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) CSRA -0.156 (0.02) *** -0.164 (0.02) *** 0.021 (0.03) BIG4 -0.080 (0.02) *** -0.101 (0.02) *** -0.054 (0.03) ROA -0.104 (0.05) -0.118 (0.11) -0.080 (0.15) 0.045 (0.06) 0.172 (0.11) 0.348 (0.15) * 0.172 (0.08) * 0.230 (0.16) 0.257 (0.21) SIZE 0.029 (0.01) * 0.003 (0.02) -0.014 (0.04) -0.009 (0.01) -0.010 (0.02) -0.111 (0.03) ** -0.066 (0.02) *** -0.023 (0.03) 0.029 (0.04) SGRTH -0.018 (0.01) 0.016 (0.03) 0.021 (0.04) 0.007 (0.02) 0.081 (0.03) ** 0.062 (0.05) 0.067 (0.02) ** 0.095 (0.04) * 0.095 (0.06) CAPINT -0.002 (0.00) -0.006 (0.00) -0.006 (0.00) -0.005 (0.00) *** -0.007 (0.00) * -0.007 (0.00) 0.003 (0.00) 0.007 (0.00) 0.008 (0.01) RDINT 0.426 (0.13) ** 0.524 (0.20) * 0.528 (0.21) * 0.410 (0.14) ** 1.019 (0.22) *** 0.947 (0.28) *** 0.567 (0.22) * 0.808 (0.32) ** 1.235 (0.33) *** LEV 0.008 (0.03) -0.105 (0.05) -0.061 (0.07) 0.002 (0.03) -0.046 (0.06) -0.097 (0.08) -0.067 (0.05) -0.023 (0.08) -0.022 (0.10) BSIZE 0.008 (0.00) ** 0.011 (0.00) ** 0.010 (0.00) * 0.009 (0.00) *** 0.009 (0.00) * 0.012 (0.00) * -0.006 (0.00) -0.015 (0.00) ** -0.011 (0.01) BINDP -0.000 (0.00) -0.000 (0.00) -0.001 (0.00) -0.000 (0.00) -0.001 (0.00) -0.002 (0.00) 0.002 (0.00) *** 0.002 (0.00) 0.001 (0.00) BGEN -0.001 (0.00) * -0.003 (0.00) *** -0.004 (0.00) *** -0.005 (0.00) *** -0.006 (0.00) *** -0.008 (0.00) *** -0.007 (0.00) *** -0.009 (0.00) *** -0.008 (0.00) *** CEOD 0.006 (0.01) -0.004 (0.02) 0.009 (0.03) 0.010 (0.01) 0.023 (0.02) 0.062 (0.03) * -0.056 (0.02) ** -0.042 (0.03) -0.059 (0.03) Constant 0.076 (0.20) 0.705 (0.38) 0.958 (0.64) 0.830 (0.21) *** 1.045 (0.38) ** 2.729 (0.59) *** 1.885 (0.30 *** 1.377 (0.51) ** 0.568 (0.75) R2 0.003 0.023 0.014 0.008 0.029 0.030 0.010 0.013 0.013 Observations 30,717 13,555 7,908 30,717 13,555 7,908 30,717 13,555 7,908 Groups 4,407 2,530 1,400 4,407 2,530 1,400 4,407 2,530 1,400
ENVD = Environmental Decoupling. SOCD = Social Decoupling. GOVD = Governmental Decoupling
Table 9 presents the results of a robustness check. The rationale for this test is that results should be similar if disclosure and performance are rated by other agencies. In this analysis, this is partially achieved by replacing the dependent variable by the normalized “E_I_gap” as created by García-Sánchez et al. (2020) in reconstructing the measure of Hawn & Ioannou (2016). This variable is a similar measure of the CSR gap but is based solely on ASSET4 data. Again, lagged values are used for the performance scores. Prevalence of understated CSR performance increases with respect to the main data; of all observations, 8.3% can be classified as ‘overstating’ and 91.7% as ‘understating.’
Results for Model (1) show that return on assets (p < .001) and board size (p < .05) have negative and significant effects, and that both capital intensity and board gender diversity have a positive and significant effect (p < .001). A minimal positive effect of board independence (p < .05) is only found in this model. Results for Model (2) demonstrate that there is a negative effect for CSRA, while results for Model (3) indicate that there is no evidence to assume any effect of BIG4. Of the control variables, return on assets also shows a negative and significant effect in Model (2). Coefficients of both capital intensity and board gender diversity are also positive and significant in Models (2) and (3). This test, therefore, partially validates the findings of the main test by only confirming a negative relationship between the general adoption of external assurance and CSR decoupling.
Table 9. Robustness check
Variable Exp. Sign (1) (2) (3) CSRA ( – ) -0.145 *** (0.02) BIG4 ( – ) -0.003 (0.02) ROA ( – ) -0.263 *** (0.07) -0.366 ** (0.12) -0.062 (0.13) SIZE ( – ) -0.028 (0.02) -0.007 (0.03) -0.058 (0.03) SGRTH ( – ) -0.023 (0.02) -0.020 (0.03) 0.005 (0.04) CAPINT ( – ) 0.006 *** (0.00) 0.010 ** (0.00) 0.009 ** (0.00) RDINT ( – ) -0.167 (0.17) 0.063 (0.22) -0.190 (0.25) LEV ( – ) -0.006 (0.04) -0.047 (0.06) 0.023 (0.06) BSIZE ( + ) -0.005 * (0.00) -0.001 (0.00) 0.005 (0.00) BINDP ( – ) 0.001 * (0.00) 0.000 (0.00) -0.001 (0.00) BGEN ( – ) 0.007 *** (0.00) 0.006 *** (0.00) 0.005 *** (0.00) CEOD ( + ) -0.021 (0.01) -0.024 (0.02) -0.001 (0.02) Constant 1.980 *** (0.27) 1.401 *** (0.43) 2.032 *** (0.54) R2 0.013 0.015 0.012 Observations 30,475 13,402 7,869 Groups 4,394 2,493 1,393
