Does good governance mean better corporate social performance? A comparative study of OECD countries

Abstract

Policymakers are increasingly concerned with enabling sustainable corporate behavior as part of the transition toward a more sustainable society. Yet, it remains unclear to what extent a country’s governance quality impacts on the sustainable behavior of firms. Drawing on the good governance literature, we hypothesize that the quality of different governance dimensions at the nation-level affects corporate social performance (CSP). This reasoning is tested using a longitudinal data set that combines nation-level and firm-level observations. Different model specifications confirm that countries with a better governance quality have higher CSP. In particular, higher levels of voice and accountability, regulatory quality, rule of law, and control of corruption are positively associated with CSP across model specifications. We conclude with a discussion of the implications for research and practice.

Introduction

Given their large impact on our economy and society, corporate behavior is a crucial factor for policymakers concerned with achieving ambitious sustainability goals (Hartmann and Uhlenbruck 2015; Koppenjan and Enserink 2009; Walker and Hills 2012). Problematically, existing research on sustainability in the public administration and public policy literatures is fragmented (Armstrong and Kamieniecki 2019), and has mostly looked at how sustainable development initiatives can best be governed at the local or city level (Krause, Feiock, and Hawkins 2016; Feiock, Krause, and Hawkins 2017; Kerret and Menahem 2016). In contrast, sustainability research in the (international) business literature focuses mainly on either the effects of corporate sustainability on financial returns, or the impact of shareholder engagement on corporate sustainability (Semenova and Hassel 2019; Dyck et al. 2019). As such, this literature largely ignores the role of nation-level institutions and governance (Ioannou and Serafeim 2012; Liang and Renneboog 2017).

In this study, we aim to connect research on the quality of governance to sustainability performance in general, and corporate social performance (CSP), in particular. Drawing on the good governance literature, we argue that different aspects of governance quality at the nation-level can affect firm-level sustainability performance. In line with Ioannou and Serafeim (2012), we conceptualize CSP as the combination of environmental sustainability and social sustainability. Environmental sustainability ensures the long-term stability and resilience of the ecosystems that support human life, whereas social sustainability facilitates the respect and promotion of human rights and other basic social rights (Sjåfjell and Bruner 2019).

We capture good governance by six governance dimensions that are often used in the literature. These are voice and accountability (VA), political stability (PS), government effectiveness (GE), regulatory quality (RQ), rule of law (RL), and control of corruption (CC). We start from the notion that the quality of these good governance dimensions at the nation-level is positively associated with CSP. For example, countries with high levels of VA enable firms to meaningfully participate in sustainability initiatives. Similarly, an effective CC makes it less likely that resources aimed at promoting sustainable corporate behavior are diverted to ineffective uses, whereas high-quality regulation can act as a stimulant for firms to act in a sustainable manner.

We test the association between good governance and CSP with a sample of listed firms incorporated in OECD countries for which CSP data are available. We focus on OECD countries for two reasons. First, although many non-OECD countries are primarily interested in governance structures that can help reduce poverty and improve economic development, sustainable development is a particularly relevant topic for (highly) developed countries (Grindle 2004; Liang and Renneboog 2017; Loorbach 2010). Second, CSP data for firms from developing countries are often missing as a result of poor data quality and a limited number of listed companies to begin with.

By combining data at the nation-level on governance characteristics with firm-level performance data, we are able empirically to study the relationship between nation-level governance and CSP across OECD countries. This study provides valuable insights into how the underlying dimensions of good governance can inform sustainable development policies. We also answer calls for more comparative cross-country research in public administration and public policy (Lee and Whitford 2009; Bouckaert 2016), and research on the relationship between nation-level institutions and CSP (Campbell 2007; Matten and Moon 2008).

In the following section, we introduce our hypotheses linking research on good governance to CSP. We then present the data and methods, followed by the Results section. We conclude with a discussion of the study’s contributions, limitations, and avenues for future research.

Good governance and CSP

The international donor community developed the good governance concept in the 1980s and 1990s with a leading role for the World Bank (Doornbos 2001). In a nutshell, good governance captures “what many would consider sensible and attractive characteristics of government” (Andrews 2008:379), such as business-friendly policies, a strong RL, and contested elections (Brinkerhoff and Brinkerhoff 2011). Yet, there is still no consensus in the literature about what good governance entails exactly, both theoretically and empirically (Andrews 2008; Langbein and Knack 2010; Nguyen et al. 2019). What is clear, however, is that donor countries have used the good governance concept to make international aid conditional on institutional reform by the aid recipient, or to give aid to countries that already meet certain governance standards (Doornbos 2001; Santiso 2001; Ear 2007). More generally, governance scores and rankings such as those put forward by the World Bank have been instrumental in comparing governance quality across and among developed and developing countries—both in academic research and in policy-making circles.

A rich empirical literature has developed over time, linking the quality of governance to different outcomes at the national level. The previous studies have shown that good governance is positively associated with economic growth (Dollar and Kraay 2003), financial development (Law and Azman-Saini 2012), foreign direct investment (Daude and Stein 2007), satisfaction with democracy (Wagner, Schneider and Halla 2009), and happiness in nations (Ott 2010), to mention just a few examples. Good governance likely also plays part in the context of CSP, as the absence of high-quality institutions and governance structures means “firms will have interests and incentives that may cause them to behave in socially irresponsible ways” (Campbell 2007:951).

A number of studies in the international business literature have investigated the relationship between nation-level institutions and different CSP dimensions. Ioannou and Serafeim (2012) use a sample of firms from 42 countries to show that strong political, economic, and cultural institutions positively affect CSP. In another comparative study, Cahan et al. (2016) inter alia find a positive association between strong nation-level institutions and corporate social responsibility disclosures. In contrast, Jackson and Apostolakou (2010) find for a sample of European firms that in liberal economies voluntary CSP activities substitute for, rather than complement, institutionalized stakeholder participation. Finally, Liang and Renneboog (2017) study the foundations of CSP using a sample of firms from 114 countries. The authors include CC and RQ in their analyses, but find no statistically significant effects of these governance dimensions on CSP. The mixed findings from this nascent research field imply that more theoretical and empirical work is required to better understand the relationship between nation-level governance and CSP.

We add to existing research by connecting the literatures on good governance and CSP. Specifically, good governance is captured by six dimensions, namely i) VA; ii) PS; iii) GE; iv) RQ; v) RL, and; vi) CC. These governance dimensions capture a broad set of institutional qualities that can meaningfully inform empirical research on governance and CSP. Furthermore, given their widespread use in practice, research on these specific good governance dimensions can inform current policy debates about effective means to pursue sustainable development goals. We hypothesize the relationship between each of the six governance dimensions and CSP below.

Voice and accountability

In the good governance literature, VA capture “the degree to which individuals are able to express their views and demand action of those in power—either through the courts, the ballot box, the media, or other institutional mechanisms” (Oh and Oetzel 2011:663). We expect VA positively to affect CSP for at least three reasons. First, much of the literature on governing sustainable development stresses the importance of collaboration between government and societal actors (Kemp, Parto, and Gibson 2005; Lim and Tsutsui 2012). Policymakers can benefit from specific knowledge by giving voice to citizens, corporations, and other relevant stakeholders (Leuenberger 2006; van Huijstee et al. 2007). Giving voice to stakeholders in the development process also increases support during the implementation phase of sustainable development initiatives (Wang et al. 2012).

Second, citizens can use elections to vote for political parties that show a greater commitment to sustainable development when there are high levels of VA. Political parties with a stronger pro-environment focus can take (legislative) action to stimulate better CSP (Hess 2014) and in so doing enhance sustainable development. For example, in the United States, the Republican Party has been found to be substantially less pro-environment than the Democratic Party (Dunlap, Xiao, and McCright 2001; McCright, Xiao, and Dunlap 2014). As elections in the United States are relatively free and fair, citizens have the option of voting for the Democratic Party in future elections if environmental protection becomes a more salient societal concern, ceteris paribus. In other countries with low VA scores, the possibility for citizens to vote for political parties with a stronger focus on sustainable development is either limited or absent altogether (Enserink and Koppenjan 2007; Mol 2009).

Third, the media play an important role when holding those in power accountable for their actions (Lio and Liu 2008). Credible media outlets are common in countries with high VA. These outlets can act as an information channel between firms and their stakeholders, and pressure firms to act more responsibly (Weaver, Klebe Treviño, and Cochran 1999). Furthermore, stakeholders are likely more demanding in terms of CSP in countries characterized by strong political accountability and media freedom (Cahan et al. 2016). Prior research has confirmed that a positive association exists between freedom of the media and different dimensions of CSP (Hartmann and Uhlenbruck 2015). Jointly, these VA mechanisms are likely to have a positive effect on CSP, which leads to our first hypothesis.

Hypothesis 1: There will be a positive association between a country’s level of voice and accountability, and CSP.

Political stability

Political stability captures “the likelihood that the government will be destabilized or overthrown by unconstitutional or violent means, including politically motivated violence and terrorism” (Kaufmann, Kraay, and Mastruzzi, 2011:223). In the literature, scholars often focus on the mirror image of PS, political instability. Political instability is strongly associated with uncertainty in the politico-economic environment (Alesina and Perotti 1996), which in turn negatively affects a country’s economic development (Daude and Stein 2007; Roe and Siegel 2011).

For firms, political (in)stability is part of the external environment, which is reflected in their risk of doing business (Gani 2007). If countries are politically unstable—and the associated risk of doing business is therefore high—firms can decide to invest less in new projects, or not invest at all (Waguespack, Birnir, and Schroeder 2005; Koppenjan and Enserink 2009). Sustainability initiatives can be costly and may not entail clear economic benefits for the firm to begin with (Orlitzky et al. 2017; Husted and de Sousa-Filho 2017). Indeed, shareholders may be unwilling to invest in sustainability projects in politically unstable countries if the expected return is not high enough (Asiedu 2002). This leads to our second hypothesis.

Hypothesis 2: There will be a positive association between a country’s level of political stability and CSP.

Government effectiveness

Government effectiveness captures how well a government is able to formulate and implement sound policies, as well as deliver public goods and services (Lio and Liu 2008). Effective governments show credible commitment to their policies, have low levels of red tape, and ensure that their civil service can operate independently from political pressures (Kaufmann et al. 2011; Arndt 2008). Government effectiveness likely affects CSP in a number of ways. First, the public sector itself is increasingly concerned with sustainable procurement (Walker and Brammer 2009). Sustainable public procurement means that governments take on a dual role by “participating in the market as purchaser and at the same time regulating it through the use of its purchasing power to advance conceptions of social justice” (McCrudden 2004:257). Governments can choose to procure goods and services from firms with a strong sustainability performance record, and hence creating economic incentives for corporations to act in a sustainable way (Brammer and Walker 2011; Eisner 2004; Walker et al. 2012). Commitment to sustainable public procurement is more pronounced in developed countries with more effective governments (Stevens 2010). For example, Walker and Phillips (2009:53) mention the viewpoint that “sustainable procurement is a Western or Northern Hemisphere concern. If you live in a developing country and are struggling to put food on the table, the health and safety conditions in your factory or the environmental harm done in the manufacturing process are pretty irrelevant.” As a result, GE is expected to positively have an impact on CSP.

Second, ineffective governments characterized by high levels of red tape and excessive bureaucracy act as a barrier for corporate sustainable development initiatives (Arevalo and Aravind 2011). Red tape creates opportunities for rent seeking and corruption (Duvanova 2014), increases transaction costs (Keyworth 2006), and harms firm performance, in general (De Jong and van Witteloostuijn 2015). As such, the presence of red tape means that company resources are wasted or misallocated, rather than used to improve, among others, CSP. To illustrate, Bozeman and DeHart-Davis (1999) study red tape in the context of the implementation of Title V of the 1990 Clean Air Act Amendments in the United States. The authors find that “one company invested several months documenting information on its insignificant emissions sources, only to have EPAs first white paper relegate documentation of such sources to a checklist in the permit application. The engineer working on the permit estimated she could have saved one-third to one-half of her time on the permit had check listing of such sources been the original policy” (Bozeman and DeHart-Davis 1999:171). Hence, we expect that CSP will be higher in countries with a more effective government.

Hypothesis 3: There will be a positive association between a country’s level of government effectiveness and CSP.

Regulatory quality

Regulatory quality captures to what extent government rules and regulations are sound. High-quality regulation signals a climate conducive for capital investment and market entry (Bertelli and Whitford 2009), which results in better economic performance, including CSP. Indeed, previous research has shown that corporate financial performance and sustainability performance can go hand in hand (Dowell, Hart, and Yeung 2000; Waddock and Graves 1997). The quality of regulation is particularly important in the context of CSP, as managers may not change their irresponsible behavior if there are no clear legal requirements to do so (Fineman and Clarke 1996; Orlitzky et al. 2017). In fact, a country with high-quality environmental regulations can even stimulate corporations “to meet and even exceed regulatory prescriptions” (Hartmann and Uhlenbruck 2015:732).

The previous research has shown a positive association between RQ and different dimensions of sustainable development. Kirkpatrick and Parker (2004) argue that improving RQ through regulatory impact assessments can positively affect sustainability performance in developing countries. Similarly, Mavragani, Nikolaou, and Tsagarakis (2016) find that RQ is positively associated with a country’s environmental performance. Scholars in the domain of environmental regulation and economic performance have also argued that the stringency of regulations is indicative of RQ (Esty and Porter 2002; Ambec et al. 2013), as stringent regulations are more likely than lax regulations to achieve regulatory environmental goals and high levels of CSP. First, stringent regulations can create incentives for firms to operate in a sustainable way. In this light, Sauvage (2014) finds that countries with more stringent environmental regulations specialize more in environmental products (e.g., products aimed at preventing and managing pollution), which in turn supports national and global pollution reduction efforts. Second, the absence of stringent regulations is expected to negatively affect CSP. Notably, different jurisdictions may attempt to lower the stringency of their environmental regulations to attract or maintain firms, which is known as a race to the bottom (Woods 2006). For example, Xing and Kolstad (2002) show that lax environmental policy in a foreign host country tends to attract more foreign direct investment from the United States for pollution intensive industries as opposed to less polluting industries. As such, CSP is likely lower for countries with less stringent regulations. This leads to the following hypothesis:

Hypothesis 4: There will be a positive association between a country’s level of regulatory quality and CSP.

Rule of law

Rule of law captures “the extent to which agents have confidence in and abide by the rules of society” (Kaufmann et al. 2011:223). Property rights are respected and protected, the perceived incidences of violent and nonviolent crime are low, contracts are enforced effectively, and confidence in the judiciary is high in societies with a strong RL (Licht, Goldschmidt, and Schwartz 2007; Lio and Liu 2008). As a result, RL plays part in investment and innovation decisions (Dollar and Kraay 2003). The previous studies have indeed shown that RL positively affects foreign direct investment (Gani 2007) and firm growth (Claessens and Laeven 2003).

The protection of property rights likely affects CSP as part of a country’s RL. Property can be defined as “a right to a benefit stream that is only as secure as the duty of all others to respect the conditions that protect that stream” (Bromley and Cernea 1989:5). Inadequate protection of property rights implies a lower rate of return for investors (Svensson 1998). Consequently, entrepreneurs and corporations are less motivated to invest in opportunity discovery (Foss and Foss 2008) and innovative activities (Lin, Lin, and Song 2010). More specifically, the absence of a credible revenue stream owing to poor protection of property rights conflicts with the notion of CSP as a means to “make a profit and make the world a better place at the same time” (Falck and Heblich 2007:247). Put differently, firms will be more inclined to improve their CSP if the future revenues of CSP improvements are protected by law.

Generally speaking, a country with a strong RL does not only lay down the rules of the game, but also enforces these rules effectively (La Porta et al. 2000). In their comparative study on environmental protection and economic growth, Castiglione, Infante, and Smirnova (2015) use a vector autoregression model to test the relationships between a country’s GDP, pollution, and RL. The authors find that effective rule enforcement is associated with less carbon dioxide emissions, whereas the opposite causal relation does not hold. Similarly, both Zhan et al. (2014) and Liu et al. (2018) find that environmental protection in China is lacking because of the country’s weak enforcement of rules and regulations. As such, scholars have suggested that improving existing regulatory frameworks and rule enforcement effectiveness are promising paths for stimulating sustainable development (de Abreu 2009; Tosun and Knill 2009). We expect a similar dynamic to apply in the context of CSP. Countries with effective rule enforcement signal to corporations that illegal unsustainable behavior will likely be observed and penalized. In contrast, a lack of rule enforcement implies that such behavior may well go unpunished. This leads to our fifth hypothesis.

Hypothesis 5: There will be a positive association between a country’s level of rule of law and CSP.

Control of corruption

Corruption, which is often defined as the abuse of public power for private gain, as well as state capture (Kaufmann et al. 2011; Hellman, Jones, and Kaufmann 2003), plays an important role in the good governance literature. Many empirical studies suggest that a country’s level of corruption is affected by a range of nation-level characteristics, such as economic development, religion, and cultural climate (Svensson 2005; Licht et al. 2007; Su and Ni 2018). Corruption has been found to have a negative impact on economic growth (Mauro 1995; Meon and Sekkat 2005), and the level of foreign direct investment (Egger and Winner 2006). Furthermore, more corrupt countries are associated with higher levels of red tape (Kaufmann, Hooghiemstra and Feeney 2018) and lower GE (Brewer, Choi, and Walker 2007).

Higher levels of corruption are likely to harm CSP, for at least two reasons (Ioannou and Serafeim 2012). First, corruption opens up opportunities for firms to engage in unethical behavior to lower costs (child labor) or increase revenues (gaining market access by paying bribes). Second, “the resulting benefits to ethical firms may be lower in more corrupt countries, since in such countries the state is less likely to provide incentives for them to be socially responsible in the form of tax exemptions, financial support, and improved infrastructure” (Ioannou and Serafeim 2012:840). This leads to the following hypothesis:

Hypothesis 6: There will be a positive association between a country’s level of control of corruption and CSP.

Data and methods

Data set and variables

We use an unbalanced panel data set with nation-level and firm-level data to explore the relationship between good governance and CSP. Our dependent variable, the CSP score, is taken from the Thomson Reuters ESG database. This database contains firm-level scores that are linked to environmental, social, and corporate governance pillars. The database “specializes in providing objective, relevant, auditable, and systematic ESG information and investment analysis tools to professional investors” (Ioannou and Serafeim 2012:844). Scores are based on publicly available information sources, including annual reports and stock exchange filings. As our research focuses on CSP, which does not reflect corporate governance performance, we include only the scores related to environmental (“Resource Use Score,” “Emissions Score,” and “Environmental Innovation Score”) and social sustainability (“Workforce score,” “Human Rights score,” “Community Score,” and “Product Responsibility Score”) in our analysis. Accordingly, our dependent variable, the CSP score, is the equally weighted average of the environmental sustainability performance score and the social sustainability performance score (Ioannou and Serafeim 2012; Dyck et al. 2019).

We use the six World Bank’s worldwide governance indicators (WGI) as independent variables. The WGI provides data on VA, PS, GE, RQ, RL, and CC. The indicators are aggregate measures based on “several hundred variables obtained from 31 different data sources, capturing governance perceptions as reported by survey respondents, nongovernmental organizations, commercial business information providers, and public sector organizations worldwide” (Kaufmann et al. 2011:221). Values for the indicators are between –2.5 and 2.5 with a mean of zero. The WGI has been used in the previous studies to investigate, among others, GE in Asia (Brewer et al. 2007), civil service structure and corruption (Rubin and Whitford 2008), country size and government performance (Jugl 2019), and aid effectiveness (Ear 2007).

We use different control variables at the firm and national level. At the firm level, we include financial structure variables (taken from the Thomson Reuters database) that are commonly used in CSP studies. Firm market value is included as an indicator of firm size, as larger firms are expected to have better CSP (Liang and Renneboog 2017; Ioannou and Serafeim 2012). Furthermore, we include the leverage ratio, which captures a firm’s relative level of debt (MacKenzie and Rees 2011; Barnea and Rubin 2010). In addition to financial structure, ownership can also vary greatly between firms, and the current literature often uses share ownership indicators to explain CSP (Barnea and Rubin 2010; Dam and Scholtens 2012). Following Jensen and Meckling (1976), any deviation from financial value maximization (such as CSP expenditure) declines when management ownership increases (Barnea and Rubin 2010). We therefore also control for the percentage of shares that are held in free float as a proxy for blockholder and insider ownership (Barnea and Rubin 2010). At the nation-level, we control for GDP per capita (taken from the World Bank database) and the legal tradition of the countries’ legal system (taken from the CIA Factbook; Liang and Renenboog 2017).

Summary statistics

Our sampling period covers a 10-year period, starting in 2008 and ending in 2017. We retrieved available CSP information for companies that are incorporated in one of the OECD countries at the time of data collection (2019). Generally speaking, companies can be assigned to countries by using either the location of the firm’s headquarters, or the firm’s place of incorporation. In this study, we have opted for the latter approach, as the country of incorporation plays a key role in terms of the government rules and regulations a firm needs to follow (Cicon et al. 2012). This approach is in line with similar comparative studies (Ioannou and Serafeim 2012). Furthermore, for most of our sample firms (>98%) the country of incorporation is the same as the country in which the firm has its headquarters. For the limited number of cases where information regarding the firm’s country of incorporation was lacking, we used the location of the firm’s headquarters to assign firms to countries.

We retrieved CSP data for 4,444 unique firms from 30 OECD countries, resulting in 28,446 firm-year observations. Table 1 summarizes the statistics for our variables.

Table 1. Full sample summary statistics.

Variable name	Mean	s.d.	Min	Max	N
Country-level data
VA	1.17	0.27	–0.71	1.74	28,446
PS	0.66	0.43	–2.01	1.59	28,446
GE	1.52	0.31	–0.03	2.25	28,446
RQ	1.45	0.32	0.04	2.09	28,446
RL	1.55	0.37	–0.67	2.10	28,446
CC	1.49	0.48	–0.93	2.45	28,446
Legal family	0.39	0.49	0	1	28,446
GDPc	3.78e + 9	4.36e + 10	7,792.25	5.45e + 11	28,446
Firm-level data
CSP	53.00	20.16	5.50	98.65	28,446
Env	53.22	23.32	2.49	99.43	28,446
Soc	52.78	20.36	4.14	99.09	28,446
Market value	1.21e + 10	2.98e + 10	3,188,720	8.69e + 11	28,446
Leverage	0.60	0.24	0.00	4.99	28,446
Free float	82.94	21.18	0	100	28,446

The summary statistics for the dependent variable CSP and its underlying Environmental and Social pillars, as well as summary statistics for the good governance scores can be found in Tables A1 and A2.

Table 2. Pooled OLS regression models with year and industry FE.

	Full sample
	(1)	(2)	(3)	(4)	(5)	(6)
	CSP	CSP	CSP	CSP	CSP	CSP
VA	11.02***
	(1.10)
PS		0.19
		(0.68)
GE			1.53
			(0.98)
RQ				8.82***
				(0.95)
RL					4.41***
					(0.88)
CC						3.63***
						(0.61)
Legal Family	6.71***	5.88***	6.01***	8.35***	7.06***	6.28***
	(0.61)	(0.64)	(0.62)	(0.63)	(0.64)	(0.62)
log(GDPc)	–0.78***	–0.60***	–0.56***	–0.63***	–0.58***	–0.59***
	(0.18)	(0.17)	(0.17)	(0.18)	(0.17)	(0.17)
log(MV)	7.53***	7.39***	7.39***	7.53***	7.41***	7.47***
	(0.18)	(0.18)	(0.18)	(0.18)	(0.18)	(0.179)
Freefl	0.03**	0.05***	0.04***	0.04***	0.03**	0.03**
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Lev	11.94***	11.44***	11.55***	12.18***	11.71***	12.02***
	(1.20)	(1.21)	(1.21)	(1.20)	(1.20)	(1.21)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Sector FE	Yes	Yes	Yes	Yes	Yes	Yes
_cons	–133.40***	–119.98***	–122.33***	–136.31***	–126.88***	–126.75***
	(4.72)	(4.59)	(4.91)	(4.93)	(4.83)	(4.77)
N	28,446	28,446	28,446	28,446	28,446	28,446
adj. R^2	0.33	0.31	0.31	0.33	0.32	0.32

The reported models are estimated using pooled OLS. All models control for year and industry fixed effects. *, **, and *** indicate statistical significance at the 10, 5, and 1% levels, respectively. Standard errors are clustered at the firm level and reported in parentheses.

Correlation matrix

Table A3 summarizes that the good governance indicators are highly correlated. Many correlations exceed 50%, and some correlations are almost as high as 90%. This is in line with other studies using the WGI data (Daude and Stein 2007; Oh and Oetzel 2011). Therefore, we conduct separate regression models for each of the good governance indicators. To determine if correlations between any of our other independent and control variables are problematically high, we examined the variance inflation factor scores. These scores, which are calculated for each regression model with one particular good governance indicator at a time, are all <2.0, which is well below the standard threshold for multicollinearity.

Results

Pooled OLS regression analysis

As we are interested in analyzing the effects of different levels of governance quality at the nation-level on CSP, we first ran pooled OLS regression models with industry and year fixed effects (Ioannou and Serafeim 2012). Standard errors are clustered at the firm level. Table 2 summarizes the results.

Table 2 shows summarizes that four out of the six good governance indicators (“CC,” “VA,”, “RQ,” and “RL” have statistically significant positive effects on CSP. “GE” and “PS” are not significant in our full sample model specification. Based on these OLS models with time and industry fixed effects, we can generally conclude that good governance at the nation-level contributes to CSP at the firm level. Results for our control variables are similar to the previous studies (Liang and Renneboog 2017; Ioannou and Serafeim 2012; MacKenzie and Rees 2011). We find that in civil law jurisdictions firms show, ceteris paribus, higher CSP. Liang and Renneboog (2017:855) explain this effect, arguing that civil law countries are “associated with state intervention in economic life through rules and regulations,” and a stakeholder orientation, which in turn improves CSP. Furthermore, the models show that larger firms seem to have higher CSP. Free float and leverage positively affect CSP in all our models. Surprisingly, we find a statistically significant negative relationship between GDP per capita and CSP for the pooled OLS model. From a methodological perspective, this result may be caused by the pooled OLS model specification (Liang and Renneboog 2017; also see the discussion below). From a theoretical perspective, the negative relationship between GDP per capita and CSP can possibly be explained through the concept of institutional voids (Doh et al. 2017), which assumes that firms may compensate for institutional shortcomings by improving their CSP. Tentatively, it stands to reason that countries with lower GDP per capita exhibit more pronounced institutional shortcomings.

Within-between random effects analysis

The pooled OLS results indicate a statistically significant positive association between four of the good governance dimensions at the nation-level, and CSP at the firm level. These findings provide support for most of our hypotheses. Yet, previous research has shown that the results obtained from analyzing this type of panel data may be driven in part by model specification. Notably, Ioannou and Serafeim (2012) find a positive statistically significant effect of the control of corruption variable on CSP in a pooled OLS model, but this relationship is no longer statistically significant when a random effects generalized least squares (GLS) model is used instead (Liang and Renneboog 2017). Similarly, Wilson and Butler (2007) replicate eight articles in top political science journals and find that alternative model specifications can lead to very different conclusions. As a result, we need to carefully consider alternative model specifications.

Generally speaking, pooled OLS models are commended for their simplicity (Bartels 2008). Pooled OLS models are used regularly to analyze time series and cross-sectional data, especially with panel-corrected standard errors (Beck and Katz 1995). As the explanatory variables of interest (good governance indicators) and the dependent variable (CSP) in our study are continuous, pooled OLS estimation is appropriate for our data (Jugl 2019). At the same time, a random effects GLS model allows the researcher to control for unobserved heterogeneity at the panel level, and thus reduce omitted variable bias. One way to determine if a random effects GLS model is appropriate to use the Breusch and Pagan Lagrangian multiplier. The results from this test show that there are indeed panel effects in our data, which we need to consider.

The distinction between pooled OLS and random effects models is only part of the story, as there can be separate within-cluster and between-cluster (or cross-sectional) effects of good governance dimensions on CSP. Indeed, the diverging results from the studies by Ioannou and Serafeim (2012) and Liang and Renneboog (2017) regarding the effect of control of corruption on CSP hint at the importance of within-cluster and between-cluster effects. As the random effects estimator combines two effects into a single effect, the within-cluster and between-cluster effects may actually cancel each other out, which could explain the lack of statistical significance of the control of corruption variable in the Liang and Renneboog (2017) study. This is a general problem with this type of analysis when within-cluster and between-cluster effects move in opposite directions (Bartels 2008).

We use the methodology put forward by Bartels (2008) to distinguish these two effects (also see Bell and Jones 2015; Jugl 2019; Jindra and Vaz 2019) using the within-between random effects (REWB) estimators. More specifically, we calculate i) the cluster-specific means of the independent variables (between-cluster variable) denoted as ${\overline{X}}_{ij}$ and ii) the within-cluster operationalization of$X_{ij},$ which can be denoted as: $X_{ij}^w= X_{ij}- {\overline{X}}_{ij}.$ It should be noted that, in line with our theoretical expectations, we are interested in the between-cluster effects that reflect different levels of governance quality across countries—while also taking into account the within-cluster effects. To control for within-panel (serial) correlation and cross-sectional heteroscedasticity, we cluster the standard errors on the panel variable (in our case, firms, see Wooldridge 2013). Although the Lagrange-multiplier (Wooldridge) test (Wooldridge 2002, also see Drukker 2003) shows that serial correlation is present in our panels, clustering the standard errors solves this issue as our data set contains panels with a large number of firms and a small number of time periods (Wooldridge 2002; Cameron and Trivedi 2005).

The results summarized in Table 3 show that the specification of the model indeed matters. Furthermore, the distinct differences in within- and between-effects illustrate the usefulness of separating these two effects when dealing with sluggish (slow to change) variables (Jugl 2019). We find a positive statistically significant between-panel effect of all good governance indicators on CSP (“_between”). These findings imply that, generally speaking, firms in countries with higher good governance scores have higher CSP scores than firms in countries with lower good governance scores. In other words, we find a cross-sectional level effect of good governance on CSP. This finding is generally in line with our pooled OLS results.

Table 3. REWB models with random intercept and year and industry FE.

	Full sample
	(1)	(2)	(3)	(4)	(5)	(6)
	CSP	CSP	CSP	CSP	CSP	CSP
VA_within	–9.63***
	(1.11)
VA_between	14.50***
	(0.96)
PS_within		–3.50***
		(0.42)
PS_between		2.11***
		(0.63)
GE_within			–5.32***
			(0.54)
GE_between			5.23***
			(0.92)
RQ_within				–2.37***
				(0.50)
RQ_between				11.68***
				(1.00)
RL_within					–5.66***
					(0.72)
RL_between					5.38***
					(0.79)
CC_within						–4.65***
						(0.52)
CC_between						5.99***
						(0.55)
Legal Family	8.45***	6.70***	7.93***	10.85***	8.87***	8.34***
	(0.58)	(0.62)	(0.60)	(0.65)	(0.63)	(0.59)
Log(GDPc)_with	5.02***	4.08***	3.00***	4.94***	4.17***	5.18***
	(0.58)	(0.59)	(0.60)	(0.59)	(0.58)	(0.58)
Log(GDPc)_between	–0.88***	–0.75***	–0.59***	–0.68***	–0.65***	–0.64***
	(0.17)	(0.18)	(0.17)	(0.17)	(0.17)	(0.17)
log(MV)_with	1.80***	1.83***	1.87***	1.84***	1.84***	1.86***
	(0.12)	(0.12)	(0.12)	(0.12)	(0.12)	(0.12)
log(MV)_between	8.36***	8.45***	8.36***	8.73***	8.46***	8.43***
	(0.20)	(0.21)	(0.21)	(0.21)	(0.21)	(0.20)
Freefl_with	0.03***	0.04***	0.03***	0.03***	0.03***	0.03***
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Freefl_between	0.02	0.04**	0.03*	0.03**	0.03**	0.02
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Lev_within	2.63***	2.64***	2.59***	2.61***	2.63***	2.80***
	(0.58)	(0.58)	(0.58)	(0.58)	(0.58)	(0.58)
Lev_between	7.76***	7.55***	7.48***	7.96***	7.41***	7.98***
	(1.09)	(1.12)	(1.12)	(1.11)	(1.12)	(1.11)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Sector FE	Yes	Yes	Yes	Yes	Yes	Yes
_cons	–160.51***	–148.18***	–154.37***	–172.61***	–156.98***	–156.43***
	(5.11)	(5.16)	(5.31)	(5.53)	(5.32)	(5.17)
sigma_u
_cons	14.51***	14.96***	14.91***	14.76***	14.91***	14.74***
	(0.17)	(0.18)	(0.17)	(0.17)	(0.17)	(0.17)
sigma_e
_cons	6.94***	6.94***	6.93***	6.94***	6.94***	6.94***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
N	25,595	25,595	25,595	25,595	25,595	25,595

The reported panel data models are estimated using REWB with random intercept following the methodology of Bartels (2008). All models control for year and industry fixed effects. *, **, and *** indicate statistical significance at the 10, 5, and 1% levels, respectively. Standard errors are clustered at the firm level and reported in parentheses.

Surprisingly, the within-panel effect is negative for all good governance indicators. This implies that, for a given firm within a particular country, a decrease in good governance at the nation-level increases CSP at the firm level. This may indeed be the case in certain situations. For example, as president Trump considered taking the United States out of the Paris climate accord, the US private sector addressed the issue of climate change itself (Crooks 2017). Furthermore, the literature on so-called institutional voids has shown that firms may decide to compensate for institutional shortcomings at the nation-level to reduce transaction costs (Doh et al. 2017), especially in developing countries. Future research could investigate these within-panel effects in more detail, which would also require a more extensive (longitudinal) data set. We briefly return to this issue in the conclusion section.

Robustness checks

As a robustness check, we also estimate the REWB models using generalized estimating equations to measure the population average effects. The results (Table A4) are very similar to the REWB models with random intercepts. In addition, following Bartels (2008), we also include a lagged dependent variable to account for dynamics in our panel (also see Noël 2019). The results summarized in Table A5 show that dynamic effects are present in our data (the within-panel effect of the lagged dependent variable, CSP_t–1 is positive and statistically significant). The between-effects of the good governance variables are again highly comparable to the results summarized in Table 3.

Next, given the large share of US companies in our sample, we have also analyzed our data using a subset of our sample from which US firms have been excluded (18,154 firms remaining, also see Ioannou and Serafeim 2012). The results of this robustness check are reported in Tables A6 and A7. In our pooled OLS estimations, the effects of control of corruption and GE on CSP are not statistically significant, whereas for the PS indicator we find a negative statistically significant relationship. This latter finding can be explained by the fact that average CSP scores for the firms in the subset of our sample increased during our sample period, whereas the PS variable showed an opposite trend. This finding may mean that firms themselves compensate for institutional voids created by political instability to achieve better CSP (Doh et al. 2017). These trends are less pronounced for the full sample that also includes US firms. In our REWB analysis, we again find a negative relationship between PS and CSP for both the between- and the within-estimators. Government effectiveness is not statistically significant in the subset of our sample. Hence, although the full sample results support all our hypotheses, results for the sample excluding US firms provide more mixed results. These findings reiterate the importance of US firms in cross-country studies of CSP (Ioannou and Serafeim 2012).

Another concern is that firms may be active in multiple countries, thus reducing the importance of their home–country institutions. To address this issue, we provide several robustness checks in which we control for the firm’s international operations using the percentage of foreign assets of the firm’s total assets, and the percentage of foreign ownership. Table A8 summarizes the results for the REWB models of Table 3 with these additional control variables; except for PS, the good governance indicators all show statistically significant positive between-effects on CSP. Similar results are found for the pooled OLS models with these control variables (not reported, available upon request from the authors).

As additional robustness checks, we reran the analyses for the full sample using the Environmental (“Env”) and Social (“Soc”) scores as separate dependent variables. We also estimated the classical static between-effects panel data models (as opposed to the REWB analysis). The results from these robustness checks (not reported here, available upon request from the authors) are similar to those reported in this article. To conclude, from the models reported in this article, the Appendix and from those that are available upon request, it follows that the REWB model specification provides consistent results, showing positive between-effects of good governance indicators on CSP. As expected, other models (including generalized linear panel data models) that do not use the REWB model specification, but combine the within-cluster and between-cluster effects, do not show the hypothesized effects. These results are also available upon request from the authors.

Discussion and conclusion

To meet ambitious sustainable development goals, policymakers across the globe must find ways to stimulate sustainable firm behavior. Yet, it remains unclear how the quality of different nation-level governance dimensions affects CSP. Although the good governance literature suggests a positive relationship between high-quality governance structures and a variety of nation-level outcomes, the alleged positive effect of good governance has not been tested in the context of CSP. In this study, we add to the governance and sustainable development debate by investigating the association between six good governance dimensions and CSP.

We test our reasoning with a sample of 4,444 unique firms from 30 OECD countries. As previous research suggests that there may be distinct within-panel and between-panel (or cross-sectional) effects in the relationship between good governance and CSP, we used REWB estimators in addition to the more commonly used pooled OLS models. Based on our findings across panels, we can conclude that firms in OECD countries with higher good governance scores generally have higher CSP scores than firms in OECD countries with lower good governance scores. We find positive statistically significant between-effects of the six good governance dimensions on CSP in our REWB model specification. These results are consistent with our pooled OLS results, which show that CSP is generally higher in countries with higher levels of VA, RQ, RL, and control of corruption. Hence, we find support for the notion that good governance means better CSP.

This study makes several contributions to the literature. First, we add to the empirical good governance literature by linking nation-level governance to firm-level CSP. Specifically, we test the relationship between six popular good governance dimensions and CSP. In so doing, we integrate insights from the public policy and international business domains. Second, we use a combination of statistical models to verify our results and increase the reliability of our findings. Our results show that model specification matters and that different estimators lead to different results.

Our results also have implications for policymakers. In a nutshell, we show the importance of governance quality as a determinant of CSP within a set of OECD countries. Although prior research has repeatedly shown that developed countries perform better than developing countries on most nation-level outcome variables, our findings show that even within the OECD higher quality governance is associated with higher CSP. As a result, policymakers that want to stimulate the transition toward a more sustainable society should consider their country’s overall governance quality. Specifically, we conclude that it is particularly important for policymakers to enable stakeholder participation in sustainable development initiatives, as well as improve the quality of regulations, safeguard the RL, and control corruption.

This study also has a number of limitations that open up possibilities for future research. First, we have included the commonly used good governance indicators put forward by the World Bank as independent variables in our statistical analyses. As it is often the case with aggregated nation-level measures, the validity and reliability of these indicators are hotly debated in the literature (Langbein and Knack 2010; Fukuyama 2013). These empirical debates reflect more fundamental theoretical discussions about the good governance concept. As a result, future research may want to verify our findings by using alternative measures for governance quality. Furthermore, our approach can be extended to use a variety of other relevant formal and informal institutions that may affect CSP in particular, or sustainable development more generally.

Second, the focus in this study has been on OECD countries—rather than a combination of developed and developing countries—as CSP data are often missing for firms from developing countries. However, our data set could possibly be extended to include a larger set of countries. A more extensive data set could enrich our understanding of the relationship between governance quality and CSP. On the one hand, we have shown for OECD countries that good governance has a positive effect on CSP, by and large. On the other hand, research on so-called institutional voids (Doh et al. 2017) has shown that, especially in developing countries, firms may invest more rather than less in CSP if there are institutional shortcomings. As such, it would be worthwhile to consider the relationship between good governance and CSP for both developed and developing countries, keeping in mind that there are few firms in developing countries for which CSP data are readily available. It should also be noted that some corporations operate worldwide and are therefore influenced, at least to some extent, by multiple national governance frameworks. Although we control for the international activities of firms in our analysis, future research may want to take a more fine-grained approach for studying the relationship between country-level governance and CSP by creating “weighted” measures of governance quality based on a firm’s actual activities within multiple countries.

Third, there may well be other factors at the industry- or firm-level that affect CSP beyond nation-level governance. In this study, we have focused on between-cluster effects driven by different levels of governance quality across countries. These between-cluster effects explain most of the variance in CSP for our sample, but we also find evidence of within-cluster effects. For example, shareholder sustainability activism may explain increased CSP when good governance indicators remain mostly unchanged (Dyck et al. 2019; Semenova and Hassel 2019). Such additional complexity is beyond the scope of this article, but future research may want to zoom in more closely on the relationships between nation-level, industry-level, and firm-level governance.

Additional information

Notes on contributors

Wesley Kaufmann

Wesley Kaufmann ([email protected]) is an associate professor of public governance at Tilburg University and a research fellow at Arizona State University. His research focuses on good governance, rules and red tape, transparency, and citizen participation.

Anne Lafarre

Anne Lafarre ([email protected]) is an assistant professor of corporate law and corporate governance at Tilburg University. Her research focuses on corporate governance, corporate sustainability, and shareholder engagement behavior.

Notes

1 Many studies in the international business domain study corporate social responsibility. This concept is very similar to how CSP is conceptualized in existing comparative studies (Ioannou and Serafeim 2012). Therefore, we use these terms interchangeably.

2 Firm-level CSP data were unavailable for firms in Estonia, Iceland, Latvia, Lithuania, Slovakia, and Slovenia.

3 Although the hypotheses tested in this research contain associations, we also tested for Granger causality in our data set. Generally stated, a variable x is Granger-causing another variable y if past values of x are significant predictors of y even when past values of y are included in the model (Lopez and Weber 2017). We found bidirectional Granger-causal relationships between the CSP-variable and the good governance variables “VA”, “RQ,” and “GE.” We also found that the “RL” variable Granger causes CSP (unidirectional Granger-causality, also see Lopez and Weber 2017).

Appendix

Table A1. CSP scores across years, by country.

			CSP		Env		Soc
Country	Observations	Firms	Mean	s.d.	Mean	s.d.	Mean	s.d.
Australia	1,844	282	47.94	18.69	46.52	21.66	49.35	19.10
Austria	132	15	58.30	14.14	61.02	17.06	55.57	16.56
Belgium	243	28	54.00	19.05	57.74	20.46	50.25	20.60
Canada	2,147	281	48.10	18.00	46.07	21.07	50.13	17.91
Chile	207	40	41.56	19.84	43.09	23.10	40.03	19.59
Czech Republic	20	4	49.00	11.69	49.90	13.84	48.11	12.91
Denmark	233	27	57.97	15.79	60.37	17.47	55.56	18.29
Finland	365	28	64.17	14.56	69.02	17.36	59.33	15.63
France	839	102	71.87	15.11	76.89	15.67	66.85	18.00
Germany	726	102	66.09	19.41	66.29	20.81	65.89	21.13
Greece	162	18	55.85	22.56	59.30	25.23	52.41	22.87
Hungary	37	4	68.27	16.51	67.33	16.35	69.21	18.41
Ireland	324	39	55.49	19.36	54.67	21.98	56.30	19.80
Israel	147	21	46.61	20.70	44.90	22.44	48.32	22.17
Italy	376	49	61.30	21.94	61.70	24.42	60.91	21.80
Japan	3,932	431	51.66	20.36	57.26	21.87	46.06	21.73
Luxembourg	97	17	56.58	20.16	56.65	20.78	56.50	21.36
Mexico	245	41	47.37	22.16	46.50	23.84	48.24	22.94
The Netherlands	399	55	64.45	19.33	63.68	21.91	65.22	19.58
New Zealand	213	50	41.03	16.80	43.11	21.81	38.95	14.71
Norway	166	21	65.60	16.66	65.65	17.09	65.55	19.93
Poland	213	30	41.57	19.01	42.58	22.49	40.55	18.51
Portugal	73	9	74.26	11.99	74.60	14.04	73.91	13.72
South Korea	858	125	54.00	23.62	57.45	24.15	50.54	25.17
Spain	402	49	71.24	18.59	70.79	19.08	71.68	20.90
Sweden	492	66	65.90	17.11	68.25	18.67	63.55	19.04
Switzerland	648	78	57.33	20.44	59.44	20.95	55.22	23.14
Turkey	226	29	54.21	19.22	54.73	24.06	53.69	18.06
United Kingdom	2,501	343	58.96	17.07	61.25	19.12	56.67	18.83
United States	10,292	2,065	48.98	18.99	46.26	22.88	51.69	18.09
Total	28,446	4,444	53.00	20.16	53.22	23.32	52.78	20.36

This table summarizes the statistics for the dependent variable CSP and its underlying Environmental and Social pillars. The average CSP, Environmental and Social scores are 53.0, 53.2, and 52.8%, respectively. This table also shows substantial differences in average CSP scores between countries. New Zealand, Chile, and Poland have the lowest CSP on average, with scores of 41.0, 41.6, and 41.6%, respectively. Portugal has the highest average CSP score (74.3%), followed by France (71.9%), Spain (71.2%), and Hungary (68.3%).

Table A2. Good governance scores across years (countries), by country (years).

Panel A
	VA		PS		GE		RQ		RL		CC
Country	Mean	s.d.	Mean	s.d.	Mean	s.d.	Mean	s.d.	Mean	s.d.	Mean	s.d.
Australia	1.40	0.05	0.96	0.07	1.62	0.07	1.84	0.07	1.78	0.07	1.90	0.10
Austria	1.39	0.04	1.19	0.14	1.61	0.12	1.47	0.06	1.84	0.05	1.56	0.12
Belgium	1.36	0.02	0.71	0.19	1.46	0.15	1.28	0.06	1.41	0.05	1.55	0.08
Canada	1.43	0.04	1.13	0.10	1.78	0.03	1.75	0.07	1.81	0.04	1.95	0.07
Chile	1.04	0.04	0.44	0.10	1.13	0.14	1.43	0.07	1.28	0.14	1.36	0.19
Czech Republic	1.00	0.03	1.01	0.04	1.00	0.06	1.08	0.09	1.09	0.05	0.42	0.13
Denmark	1.56	0.05	0.96	0.08	1.99	0.16	1.77	0.11	1.94	0.07	2.32	0.09
Finland	1.53	0.04	1.28	0.17	2.07	0.16	1.82	0.07	1.99	0.06	2.23	0.05
France	1.21	0.05	0.37	0.24	1.42	0.07	1.17	0.07	1.45	0.03	1.39	0.08
Germany	1.38	0.04	0.79	0.12	1.64	0.09	1.64	0.12	1.68	0.08	1.81	0.04
Greece	0.75	0.11	–0.12	0.13	0.42	0.13	0.51	0.23	0.44	0.23	–0.06	0.09
Hungary	0.68	0.20	0.71	0.08	0.59	0.09	0.88	0.18	0.61	0.16	0.27	0.14
Ireland	1.32	0.03	0.98	0.09	1.44	0.10	1.69	0.11	1.69	0.12	1.61	0.09
Israel	0.68	0.07	–1.10	0.23	1.32	0.06	1.23	0.07	1.01	0.09	0.90	0.13
Italy	1.00	0.05	0.42	0.10	0.43	0.07	0.78	0.11	0.38	0.06	0.12	0.09
Japan	1.03	0.05	0.99	0.07	1.61	0.15	1.18	0.13	1.42	0.11	1.55	0.11
Luxembourg	1.56	0.05	1.39	0.06	1.69	0.04	1.71	0.07	1.81	0.06	2.07	0.06
Mexico	0.04	0.10	–0.72	0.08	0.20	0.11	0.34	0.09	–0.53	0.06	–0.60	0.21
The Netherlands	1.54	0.04	1.00	0.11	1.80	0.05	1.83	0.11	1.86	0.06	2.01	0.11
New Zealand	1.55	0.03	1.47	0.14	1.82	0.06	1.97	0.10	1.94	0.05	2.28	0.04
Norway	1.65	0.06	1.25	0.08	1.88	0.05	1.60	0.12	1.97	0.06	2.17	0.11
Poland	0.98	0.11	0.85	0.21	0.69	0.09	0.97	0.06	0.72	0.12	0.63	0.09
Portugal	1.13	0.05	0.86	0.12	1.13	0.12	0.85	0.13	1.08	0.04	0.99	0.08
South Korea	0.69	0.04	0.25	0.10	1.13	0.08	1.02	0.10	1.02	0.08	0.50	0.07
Spain	1.07	0.06	0.03	0.28	1.07	0.09	0.99	0.15	1.06	0.10	0.85	0.27
Sweden	1.59	0.05	1.07	0.08	1.89	0.09	1.80	0.10	1.97	0.04	2.22	0.06
Switzerland	1.57	0.04	1.32	0.07	1.96	0.09	1.71	0.12	1.86	0.09	2.08	0.06
Turkey	–0.31	0.21	–1.30	0.37	0.27	0.13	0.30	0.12	–0.02	0.13	–0.03	0.13
United Kingdom	1.31	0.02	0.38	0.11	1.57	0.09	1.74	0.08	1.74	0.07	1.74	0.11
United States	1.10	0.03	0.50	0.13	1.52	0.04	1.43	0.13	1.62	0.02	1.36	0.05
Total	1.17	0.27	0.66	0.43	1.52	0.31	1.45	0.32	1.55	0.37	1.49	0.48
Total (no US firms)	1.21	0.33	0.75	0.50	1.53	0.38	1.45	0.39	1.52	0.46	1.56	0.59
Panel B
Year	VA		PS		GE		RQ		RL		CC
	Mean	s.d.	Mean	s.d.	Mean	s.d.	Mean	s.d.	Mean	s.d.	Mean	s.d.
2008	1.19	0.23	0.73	0.34	1.56	0.32	1.46	0.28	1.55	0.34	1.53	0.43
2009	1.18	0.23	0.63	0.45	1.51	0.30	1.39	0.28	1.53	0.35	1.48	0.46
2010	1.17	0.24	0.63	0.39	1.53	0.31	1.40	0.32	1.53	0.37	1.48	0.51
2011	1.18	0.26	0.72	0.40	1.50	0.31	1.41	0.33	1.50	0.36	1.47	0.49
2012	1.22	0.28	0.72	0.41	1.49	0.30	1.37	0.31	1.52	0.38	1.53	0.49
2013	1.20	0.29	0.75	0.42	1.52	0.30	1.38	0.33	1.51	0.38	1.49	0.49
2014	1.16	0.29	0.71	0.43	1.54	0.34	1.41	0.37	1.61	0.42	1.50	0.52
2015	1.16	0.28	0.73	0.41	1.53	0.31	1.40	0.33	1.58	0.38	1.50	0.50
2016	1.15	0.27	0.58	0.44	1.51	0.30	1.52	0.30	1.56	0.36	1.48	0.47
2017	1.15	0.27	0.53	0.43	1.52	0.29	1.59	0.30	1.58	0.33	1.48	0.44
Total	1.17	0.27	0.66	0.43	1.52	0.31	1.45	0.32	1.55	0.37	1.49	0.48
Total (no US firms)	1.21	0.33	0.75	0.50	1.53	0.38	1.45	0.39	1.52	0.46	1.56	0.59

Panel A: Substantial differences between countries with regard to their good governance scores. For example, Turkey has the lowest average score in our sample for the dimension “PS” (–1.3), whereas New Zealand has the highest average score on this dimension (1.47). Interestingly, high good governance scores are not necessarily associated with high levels of CSP. Countries such as New Zealand, Canada, and Australia have very high governance scores, but below average CSP scores. These results are consistent with the previous studies (Ioannou and Serafeim 2012; Liang and Renneboog 2017; Dyck et al. 2019). At the same time, many countries with poor governance scores are associated with relatively low average CSP (e.g., Mexico and Poland). This is in line with our theoretical expectations. Although the standard deviations of the good governance indicators for the total sample show that there is sufficient variation, the standard deviations of the good governance indicators within countries are very small, indicating that these variables do not change very much over time. Panel B: Most of the good governance indicators are relatively stable over time.

Table A3. Correlation table.

	1.	2.	3.	4.	5.	6.	7.	8.	9.	10.	11.	12.	13.	14.
1. CSP	1.00
2. Env	0.93***	1.00
3. Soc	0.91***	0.70***	1.00
4. VA	0.07***	0.05***	0.08***	1.00
5. PS	–0.02***	0.01***	–0.05***	0.70***	1.00
6. GE	–0.01	0.00	–0.01***	0.75***	0.67***	1.00
7. RQ	–0.02***	–0.05***	0.02***	0.80***	0.45***	0.75***	1.00
8. RL	–0.01	–0.03***	0.02***	0.83***	0.59***	0.90***	0.86***	1.00
9. CC	0.02***	0.03***	0.01*	0.86***	0.70***	0.90***	0.81***	0.89***	1.00
10. Legal family	0.18***	0.25***	0.06***	–0.18***	0.14***	–0.19***	–0.47***	–0.42***	–0.17***	1.00
11. log(GDPc)	–0.06***	–0.06***	–0.05***	0.09***	0.15***	–0.08***	0.01***	–0.02***	–0.01*	0.01**	1.00
12. log(Market value)	0.52***	0.47***	0.49***	–0.08***	–0.06***	–0.03***	–0.15***	–0.07***	–0.09***	0.12***	–0.03***	1.00
13. Free float	0.03***	–0.01**	0.08***	0.16***	0.10***	0.26***	0.24***	0.31***	0.20***	–0.39***	–0.05***	0.06***	1.00
14. Leverage	0.17***	0.14***	0.17***	–0.07***	–0.12***	–0.11***	–0.11***	–0.08***	–0.12***	0.04***	0.01	0.15***	0.01**	1.00

*Significant at 10% level; **significant at 5% level; ***significant at 1% level.

Table A4. GEE REWB models with year and industry FE.

	Full sample
	(1)	(2)	(3)	(4)	(5)	(6)
	CSP	CSP	CSP	CSP	CSP	CSP
VA_within	–9.45***
	(1.08)
VA_between	14.51***
	(0.98)
PS_within		–3.47***
		(0.41)
PS_between		2.13***
		(0.65)
GE_within			–5.28***
			(0.52)
GE_between			5.24***
			(0.95)
RQ_within				–2.42***
				(0.49)
RQ_between				11.68***
				(1.03)
RL_within					–5.67***
					(0.70)
RL_between					5.38***
					(0.82)
CC_within						–4.58***
						(0.50)
CC_between						6.00***
						(0.57)
Legal family	8.47***	6.73***	7.96***	10.87***	8.90***	8.37***
	(0.60)	(0.64)	(0.62)	(0.67)	(0.65)	(0.61)
Log(GDPc)_with	4.99***	4.04***	2.98***	4.91***	4.13***	5.14***
	(0.57)	(0.56)	(0.58)	(0.57)	(0.57)	(0.57)
Log(GDPc)_betw	–0.88***	–0.75***	–0.59***	–0.68***	–0.66***	–0.64***
	(0.18)	(0.18)	(0.18)	(0.18)	(0.18)	(0.18)
log(MV)_with	1.79***	1.82***	1.86***	1.83***	1.83***	1.86***
	(0.11)	(0.11)	(0.11)	(0.11)	(0.11)	(0.11)
log(MV)_betw	8.36***	8.45***	8.36***	8.73***	8.47***	8.43***
	(0.21)	(0.21)	(0.21)	(0.21)	(0.21)	(0.21)
Freefl_with	0.03***	0.04***	0.03***	0.03***	0.03***	0.03***
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Freefl_betw	0.02	0.04**	0.03	0.03**	0.03*	0.02
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Lev_with	2.62***	2.63***	2.58***	2.59***	2.61***	2.79***
	(0.57)	(0.56)	(0.56)	(0.56)	(0.56)	(0.56)
Lev_betw	7.72***	7.49***	7.43***	7.91***	7.35***	7.92***
	(1.12)	(1.15)	(1.15)	(1.13)	(1.14)	(1.14)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Sector FE	Yes	Yes	Yes	Yes	Yes	Yes
_cons	–160.59***	–148.28***	–154.49***	–172.72***	–157.08***	–156.54***
	(5.24)	(5.31)	(5.46)	(5.67)	(5.47)	(5.33)
N	25,595	25,595	25,595	25,595	25,595	25,595

The reported panel data models are GEE (generalized estimating equation) REWB models estimated following the methodology of Bartels (2008). All models control for year and industry fixed effects. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. Standard errors are clustered at the company level and reported in parentheses.

Table A5. Dynamic REWB models with random intercept and year and industry FE.

	Full sample
	(1)	(2)	(3)	(4)	(5)	(6)
	CSP	CSP	CSP	CSP	CSP	CSP
CSPt – 1_within	0.59***	0.59***	0.59***	0.59***	0.59***	0.59***
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
VA_within	–5.16***
	(0.89)
VA_between	12.70***
	(0.91)
PS_within		–1.04***
		(0.34)
PS_between		1.53*
		(0.60)
GE_within			–2.58***
			(0.43)
GE_between			4.28***
			(0.87)
RQ_within				–1.33***
				(0.41)
RQ_between				10.68***
				(0.94)
RL_within					–2.93***
					(0.58)
RL_between					4.75***
					(0.75)
CC_within						–2.50***
						(0.42)
CC_between						5.18***
						(0.52)
Legal Family	7.45***	6.09***	6.92***	9.59***	7.80***	7.31***
	(0.55)	(0.58)	(0.56)	(0.62)	(0.59)	(0.56)
Log(GDPc)_with	2.23***	1.88***	1.25**	2.19***	1.79***	2.33***
	(0.47)	(0.47)	(0.49)	(0.47)	(0.47)	(0.47)
Log(GDPc)_betw	–0.80***	–0.68***	–0.55***	–0.62***	–0.60***	–0.59***
	(0.16)	(0.17)	(0.16)	(0.16)	(0.16)	(0.16)
log(MV)_with	1.27***	1.29***	1.31***	1.29***	1.29***	1.31***
	(0.09)	(0.09)	(0.09)	(0.09)	(0.09)	(0.09)
log(MV)_betw	7.50***	7.53***	7.46***	7.78***	7.55***	7.53***
	(0.19)	(0.20)	(0.19)	(0.19)	(0.19)	(0.19)
Freefl_with	0.01**	0.01**	0.01**	0.01*	0.01**	0.01**
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Freefl_betw	0.01	0.03	0.02*	0.02*	0.02	0.01
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Lev_with	0.98**	1.00**	0.97**	0.97**	0.98**	1.08**
	(0.46)	(0.46)	(0.46)	(0.46)	(0.46)	(0.46)
Lev_betw	7.15***	6.90***	6.93***	7.38***	6.86***	7.34***
	(1.02)	(1.05)	(1.04)	(1.03)	(1.04)	(1.04)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Sector FE	Yes	Yes	Yes	Yes	Yes	Yes
_cons	–134.40***	–122.85***	–127.96***	–144.95***	–130.55***	–130.27***
	(4.82)	(4.86)	(4.99)	(5.19)	(5.00)	(4.87)
sigma_u
_cons	13.83***	14.18***	14.13***	13.96***	14.12***	14.00***
	(0.16)	(0.16)	(0.16)	(0.16)	(0.16)	(0.16)
sigma_e
_cons	5.59***	5.59***	5.59***	5.59***	5.59***	5.59***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
N	25,595	25,595	25,595	25,595	25,595	25,595

Note: The reported dynamic panel data models are estimated using REWB with random intercept following the methodology of Bartels (2008). All models control for year and industry fixed effects. *, **, and *** indicate statistical significance at the 10, 5, and 1% levels, respectively. Standard errors are clustered at the company level and reported in parentheses. The models report the findings for the full sample. CSP_{t – 1} represents the lagged dependent variable.

Table A6. Pooled OLS regression models with year and industry FE for the subset sample.

	Sample without US companies
	(1)	(2)	(3)	(4)	(5)	(6)
	CSP	CSP	CSP	CSP	CSP	CSP
VA	6.55***
	(1.18)
PS		–2.32***
		(0.77)
GE			–0.61
			(1.03)
RQ				3.55***
				(1.15)
RL					2.67***
					(0.93)
CC						0.86
						(0.70)
Legal family	1.68**	0.40	0.17	1.94**	1.18	0.58
	(0.76)	(0.73)	(0.74)	(0.88)	(0.78)	(0.77)
log(GDPc)	–0.61***	–0.35**	–0.47***	–0.50***	–0.47***	–0.47***
	(0.17)	(0.17)	(0.17)	(0.17)	(0.17)	(0.17)
log(MV)	7.70***	7.68***	7.73***	7.69***	7.70***	7.72***
	(0.23)	(0.23)	(0.23)	(0.23)	(0.23)	(0.23)
Freefl	0.04***	0.07***	0.06***	0.05***	0.05***	0.05***
	(0.01)	(0.02)	(0.02)	(0.01)	(0.02)	(0.02)
Lev	17.89***	17.38***	17.81***	18.03***	18.06***	18.00***
	(1.89)	(1.89)	(1.89)	(1.89)	(1.90)	(1.89)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Sector FE	Yes	Yes	Yes	Yes	Yes	Yes
_cons	–133.64***	–127.43***	–127.41***	–132.50***	–131.26***	–129.03***
	(5.60)	(5.58)	(5.73)	(5.79)	(5.70)	(5.66)
N	18,154	18,154	18,154	18,154	18,154	18,154
adj. R^2	0.33	0.33	0.32	0.33	0.33	0.32

The reported models are estimated using Pooled OLS. All models control for year and industry fixed effects. *, **, and *** indicate statistical significance at the 10, 5, and 1% levels, respectively. Standard errors are clustered at the firm level and reported in parentheses.

Table A7. REWB models with random intercept and year and industry FE.

	Sample without US companies
	(1)	(2)	(3)	(4)	(5)	(6)
	CSP	CSP	CSP	CSP	CSP	CSP
VA_within	–10.73***
	(1.24)
VA_between	7.46***
	(1.11)
PS_within		–4.52***
		(0.47)
PS_between		–1.86***
		(0.71)
GE_within			–5.54***
			(0.59)
GE_between			1.09
			(0.98)
RQ_within				–3.83***
				(0.66)
RQ_between				3.88***
				(1.19)
RL_within					–6.50***
					(0.80)
RL_between					3.04***
					(0.83)
CC_within						–4.55***
						(0.57)
CC_between						1.92***
						(0.64)
Legal Family	1.71**	–0.05	0.19	1.74*	1.00	0.80
	(0.80)	(0.77)	(0.78)	(0.95)	(0.82)	(0.81)
Log(GDPc)_within	6.61***	5.35***	3.86***	6.65***	5.12***	5.61***
	(0.76)	(0.74)	(0.76)	(0.77)	(0.75)	(0.74)
Log(GDPc)_between	–0.58***	–0.35*	–0.43**	–0.46**	–0.44**	–0.45**
	(0.18)	(0.18)	(0.18)	(0.18)	(0.18)	(0.18)
log(MV)_within	1.57***	1.55***	1.66***	1.65***	1.62***	1.60***
	(0.14)	(0.14)	(0.14)	(0.14)	(0.14)	(0.14)
log(MV)_between	8.61***	8.60***	8.63***	8.71***	8.67***	8.63***
	(0.29)	(0.29)	(0.29)	(0.29)	(0.29)	(0.29)
Freefl_within	0.03***	0.04***	0.03***	0.03***	0.03***	0.03***
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Freefl_between	0.05***	0.08***	0.06***	0.06***	0.06***	0.06***
	(0.02)	(0.02)	(0.02)	(0.02)	(0.02)	(0.02)
Lev_within	3.79***	3.65***	3.58***	3.62***	3.64***	3.73***
	(0.79)	(0.79)	(0.79)	(0.79)	(0.79)	(0.79)
Lev_between	16.79***	16.83***	17.19***	17.34***	17.27***	17.25***
	(1.66)	(1.68)	(1.68)	(1.68)	(1.68)	(1.67)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Sector FE	Yes	Yes	Yes	Yes	Yes	Yes
_cons	–161.24***	–154.38***	–156.43***	–162.62***	–160.18***	–157.55***
	(6.87)	(6.88)	(6.96)	(7.17)	(6.96)	(6.90)
sigma_u
_cons	14.97***	15.13***	15.14***	15.15***	15.12***	15.11***
	(0.24)	(0.24)	(0.24)	(0.24)	(0.24)	(0.24)
sigma_e
_cons	7.09***	7.08***	7.08***	7.09***	7.09***	7.09***
	(0.04)	(0.04)	(0.04)	(0.04)	(0.04)	(0.04)
N	16,932	16,932	16,932	16,932	16,932	16,932

The reported panel data models are estimated using REWB with random intercept following the methodology of Bartels (2008). All models control for year and industry fixed effects. *, **, and *** indicate statistical significance at the 10, 5, and 1% levels, respectively. Standard errors are clustered at the company level and reported in parentheses.

Table A8. REWB models with random intercept and year and industry FE (including MNC control).

	Full sample (with MNC controls)
	(1)	(2)	(3)	(4)	(5)	(6)
	CSP	CSP	CSP	CSP	CSP	CSP
VA_within	–6.88***
	(1.33)
VA_between	12.42***
	(1.14)
PS_within		–1.95***
		(0.50)
PS_between		0.43
		(0.70)
GE_within			–3.75***
			(0.63)
GE_between			2.67**
			(1.06)
RQ_within				–1.11*
				(0.58)
RQ_between				8.84***
				(1.17)
RL_within					–3.75***
					(0.86)
RL_between					3.39***
					(0.95)
CC_within						–2.64***
						(0.63)
CC_between						4.45***
						(0.65)
ForeignAsset%_within	0.00	0.00	0.00	0.00	0.00	0.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
ForeignAsset%_between	0.10***	0.15***	0.14***	0.12***	0.14***	0.12***
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
ForeignFreefl%_within	–0.01	–0.01	–0.01	–0.01	–0.01	–0.01
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
ForeignFreefl%_between	0.08***	0.11***	0.11***	0.09***	0.11***	0.10***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
Legal Family	7.25***	6.06***	6.55***	8.90***	7.17***	6.92***
	(0.63)	(0.66)	(0.64)	(0.72)	(0.68)	(0.63)
Log(GDPc)_within	4.61***	4.09***	3.38***	4.55***	4.11***	4.76***
	(0.68)	(0.69)	(0.70)	(0.69)	(0.69)	(0.69)
Log(GDPc)_between	–0.81***	–0.69***	–0.62***	–0.66***	–0.64***	–0.63***
	(0.19)	(0.20)	(0.20)	(0.19)	(0.20)	(0.19)
log(MV)_within	1.82***	1.85***	1.88***	1.86***	1.85***	1.87***
	(0.14)	(0.13)	(0.13)	(0.13)	(0.13)	(0.13)
log(MV)_between	8.06***	7.96***	7.93***	8.24***	8.00***	8.03***
	(0.22)	(0.23)	(0.22)	(0.23)	(0.22)	(0.22)
Freefl_within	0.00	0.00	0.00	0.00	0.00	0.00
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Freefl_between	0.04**	0.06***	0.06***	0.05***	0.05***	0.05***
	(0.02)	(0.02)	(0.02)	(0.02)	(0.02)	(0.02)
Lev_within	2.57***	2.62***	2.54***	2.62***	2.58***	2.71***
	(0.66)	(0.66)	(0.66)	(0.66)	(0.66)	(0.66)
Lev_between	7.31***	6.93***	7.03***	7.47***	7.03***	7.44***
	(1.17)	(1.19)	(1.19)	(1.18)	(1.19)	(1.18)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Sector FE	Yes	Yes	Yes	Yes	Yes	Yes
_cons	–161.73***	–147.08***	–151.15***	–169.24***	–154.42***	–155.60***
	(5.65)	(5.56)	(5.82)	(6.25)	(5.90)	(5.68)
sigma_u	14.65***	14.93***	14.91***	14.82***	14.92***	14.82***
_cons	(0.18)	(0.19)	(0.19)	(0.19)	(0.19)	(0.19)
sigma_e	6.77***	6.77***	6.77***	6.77***	6.77***	6.77***
_cons	(0.04)	(0.04)	(0.04)	(0.04)	(0.04)	(0.04)
N	19,908	19,908	19,908	19,908	19,908	19,908

The reported panel data models are estimated using REWB with random intercept following the methodology of Bartels (2008). All models control for year and industry fixed effects. *, **, and *** indicate statistical significance at the 10, 5, and 1% levels, respectively. Standard errors are clustered at the company level and reported in parentheses. The variables ForeignAsset% (the percentage of the firm’s foreign assets to its total assets) and ForeignFreefl% (the percentage of strategic share holdings of 5% or more held in a country outside that of the issuer) are the Datastream time series variables with codes WC08736 and NOSHFR, respectively. Models with the percentage of foreign sales of total sales instead of the percentage of foreign assets of total assets yield very similar results (Datastream variable with code WC08731).