CSR and Developing Countries

CSR has been defined as “the responsibility of enterprises for their impacts on society” (European Commission, 2011: 6). As companies respond to the needs of the society they are embedded in, their CSR activities are likely to display a distinctly national flavour (Matten & Moon, Reference Matten and Moon2008; Wood, Reference Wood1991). Campbell (Reference Campbell2007) argues that firms are more likely to engage in CSR if there is strong and well-enforced regulation and/or an effective system of self-regulation, shored up by active civil society organizations, institutional investors and the media (see also Aguilera, Williams, Conley & Rupp, Reference Aguilera, Williams, Conley and Rupp2006). Businesses worldwide have expressed a remarkable interest in CSR and its formalized infrastructure, which ranges from codes of conduct through CSR standards to sustainability reporting along the guidelines of the Global Reporting Initiative (Lee, Reference Lee2008).

Considerable CSR research has so far been undertaken on firms from developed countries (Egri & Ralston, Reference Egri and Ralston2008). However, there are key differences in approaches to CSR between developed and developing countries. In part, such differences result from the stage of economic development a country has reached. Many developing countries face a trade-off between social development and environmental protection (Muller & Kolk, Reference Muller and Kolk2009). The regulation of corporate social and environmental impacts is also often less stringent than in developed countries (Blowfield & Frynas, Reference Blowfield and Frynas2005). Due to a limited capacity of regulators, compliance with even basic legislation cannot be taken for granted in many cases (Fox, Reference Fox2004). The political system of many developing countries is furthermore shaped by a low salience of NGOs (Logsdon, Thomas & Van Buren, Reference Logsdon, Thomas and Van Buren2006), while ethical consumerism is often not an established phenomenon either (Newell & Muro, Reference Newell and Muro2006).

At the same time, CSR is hardly a new idea for most developing countries. Albeit often termed differently, there have been similar conceptualizations of business responsibilities towards society in many countries (Blowfield & Frynas, Reference Blowfield and Frynas2005), whether these are based on the Confucian tradition in China (Wang & Juslin, Reference Wang and Juslin2009) or the social welfare tradition of the Catholic Church in Latin America (Logsdon et al., Reference Logsdon, Thomas and Van Buren2006). Thus, distinctive local forms of engagement with CSR are emerging in developing countries (Fox, Reference Fox2004).

MNEs find themselves subject to a particularly complex dynamic: Being exposed to various home and host country pressures to engage in socially responsible behaviour, they face not only increasingly complex but sometimes also competing social expectations (Arthaud-Day, Reference Arthaud-Day2005; Husted & Allen, Reference Husted and Allen2006; see also Kolk, Reference Kolk2016). This situation has a number of consequences. MNEs have become a key mechanism for the global spread of CSR as they implement CSR policies and instruments within their own operations as well as along their supply chains (Gugler & Shi, Reference Gugler and Shi2009). At the same time, MNEs may perform well on some CSR dimensions and less well on others, well in some geographic contexts and less well in others (Strike, Gao & Bansal, Reference Strike, Gao and Bansal2006). An increasing professionalisation of CSR may also lead to a concentration of CSR-related decision-making at corporate headquarters, which may de-emphasize the needs of local communities where the MNE’s subsidiaries operate (Barkemeyer & Figge, Reference Barkemeyer and Figge2014). These dynamics have so far largely been studied with regard to MNEs from developed countries. However, in view of the rise of DMNEs, we would expect similar dynamics for these firms too. We propose to study these dynamics through an analysis of DMNE codes of conduct.

Codes of Conduct

The code of conduct is one of the most widely adopted CSR tools (KPMG, 2008; Wheldon & Webley, Reference Wheldon and Webley2013) as well as being a highly visible aspect of global economic governance (Vogel, Reference Vogel2010). It has been defined as “an independent, company-specific document which delineates company responsibilities towards stakeholders and/or employee responsibilities” (Kaptein, Reference Kaptein2004: 16). Internally, a code of conduct can encourage adherence to shared ethical aspirations and compliance with organizational rules (Erwin, Reference Erwin2011) as well as introducing coherent standards across the company (Carasco & Singh, Reference Carasco and Singh2003). Externally, codes can signal trustworthiness to outside stakeholders, such as regulators, customers, suppliers or shareholders (Cressey & Moore, Reference Cressey and Moore1983; Perez-Batres, Doh, Miller & Pisani, Reference Perez-Batres, Doh, Miller and Pisani2012). However, a major criticism of codes of conduct is that these documents do not have the same clout as government regulation (Vogel, Reference Vogel2010). Scholars also noted a high degree of variability in code quality among companies, which is likely to impact code effectiveness (Erwin, Reference Erwin2011).

Codes emerged first in the United States (Cressey & Moore, Reference Cressey and Moore1983). In recent decades, these documents have spread to other nations such as Canada and Australia (Singh, Carasco, Svensson, Wood & Callaghan, Reference Singh, Carasco, Svensson, Wood and Callaghan2005), the United Kingdom (Preuss, Reference Preuss2010), France and Germany (Langlois & Schlegelmilch, Reference Langlois and Schlegelmilch1990). However, studies into the adoption of codes by companies from developing countries are still rare (Melé et al., Reference Melé, Debeljuh and Arruda2006; for exceptions see Callaghan et al., Reference Callaghan, Lee, Donmez, Aydinlik, Svensson and Wood2009; Welford, Reference Welford2005). At the same time, these firms are under pressure to adopt codes from a variety of sources, ranging from requirements of international trade agreements (e.g. Mexico due to its membership of both OECD and NAFTA [Hood & Logsdon, Reference Hood and Logsdon2002]), through demands by overseas customers and financial markets (e.g. China [Egels-Zandén, Reference Egels-Zandén2014]), to pressure for the improvement of corporate governance (e.g. South Africa [Fig, Reference Fig2005]). However, these pressures have so far been largely presented in isolation. Hence our aim—to provide a comparative global study by examining codes of DMNEs from Latin America, Africa and Asia—leads us to our first research question:

Towards a Multi-Dimensional Framework of CSR

The literature has increasingly recognized that CSR has a multi-facetted nature (Arthaud-Day, Reference Arthaud-Day2005; Wood, Reference Wood1991). As a key example of a multidimensional model, Carroll’s (Reference Carroll1979) influential pyramid of CSR encompasses a company’s economic, legal, ethical and discretionary responsibilities towards society. Developing this work further, Wood (Reference Wood1991) presented a framework of corporate social performance that combines principles of social responsibility, processes of social responsiveness and concrete outcomes of social behaviour. Arthaud-Day (Reference Arthaud-Day2005) developed a model that combines three dimensions: (1) a transnational dimension, which builds on Bartlett and Ghoshal’s (Reference Bartlett and Ghoshal2000) typology of international business strategies; (2) a content domain, which covers human rights, labour standards and the environment as suggested by the UN Global Compact; and (3) an operationalization perspective, covering what managers believe they should be doing, what society expects the firm to do and how the company operationalizes these concerns. However, such frameworks have proved difficult to operationalize (Acar, Aupperle & Lowy, Reference Acar, Aupperle and Lowy2001; Wood & Jones, Reference Wood and Jones1995). With regard to Carroll’s framework, for example, Visser (Reference Visser, Huniche and Pedersen2006) has argued that the relative priorities of CSR in many developing countries differ from the North American ordering; hence “Carroll’s CSR Pyramid may not be the best model for understanding CSR in general” (Visser, Reference Visser, Huniche and Pedersen2006: 29).

We propose a multi-dimensional framework that is more parsimonious but at the same time easier to operationalize—one that goes back to the work of Zenisek (Reference Zenisek1979). He suggests a conceptualization of CSR based on the fit between the ethic of a business and societal expectations regarding the private sector’s proper conduct. The business ethic, in turn, consists of both an operational dimension, i.e. concrete actions by the firm, and a belief system, i.e. values and principles the company and its managers espouse. In the wake of Zenisek’s (Reference Zenisek1979) work, we propose to operationalize the multi-dimensional nature of CSR as comprising three different dimensions:

• ▪ First, controversial behaviour that companies require their employees to avoid, such as not engaging in bribery. These issues are partly shaped by their national context but even more so by operational necessities, international regulation, stock market listing requirements and so forth. Hence, we would expect this dimension to show a high degree of similarity when firms from different countries are compared.

• ▪ Second, corporate commitments to society. Here, companies respond to the needs of a specific society. These are likely to be, in the first instance, requirements of the firm’s home society; however, these may overlap with requirements from other countries it operates in. Hence, we would expect this dimension to show a medium degree of similarity when firms from different countries are compared.

• ▪ Thirdly, ethical principles and values that companies commit themselves to uphold. Of the three, these are the deepest manifestation of their national context and hence likely to be less responsive to global pressure for change. Consequently, we would expect this dimension to show a low degree of similarity when firms from different countries are compared.

We explore these dimensions of CSR through an analysis of the content of codes of conduct adopted by DMNEs, which leads us to our second research question:

In summary, we focus on codes of conduct as a representation of a company’s CSR values, principles and actions. We expect code adoption rates and content to vary but with common patterns emerging among developing countries. Furthermore, we expect those differences to be visible in the three dimensions we identified on the basis of Zenisek’s (Reference Zenisek1979) framework.

Selection of Study Firms

Given the wide range of ownership patterns in many developing countries (Dam & Scholtens, Reference Dam and Scholtens2012), we sought to identify large DMNEs irrespective of ownership (i.e. covering publicly limited corporations, private companies as well as state-owned enterprises). As share indices could not be used, companies were identified through the Forbes Global 2000 Index, which applies four measures—sales, market value, assets and profits—to generate a composite measure of company size. The Forbes Global 2000 Index of 2014 included a total of 606 companies from 35 developing countries. Our classification of countries as ‘developing country’ follows that of UNCTAD (2013), which classified all OECD member countries except Chile, Mexico, South Korea and Turkey as well as a number of European non-OECD member states as ‘developed’ countries. ^{Footnote 2} Bermuda (9 companies) and Cayman Islands (1 company) were excluded due to their status as tax havens (for CSR in tax haven-based companies, see Preuss, Reference Preuss2012). Also, in order to ensure a minimum level of representativeness of the country subsamples, only countries with at least 5 codes were included in the final sample. For this reason, Bahrain (2 companies in the Forbes Global 2000), Egypt (1), Jordan (1), Kazakhstan (3), Kuwait (4), Lebanon (2), Mauritius (1), Morocco (3), Oman (1), Pakistan (2), Peru (2), Puerto Rico (1), Togo (1) Venezuela (2) and Vietnam (2) were excluded from the sample.

The resultant set of 568 companies was screened using the Bureau van Dijk Orbis and EBSCO Business databases in order to distinguish between multinational and purely domestic firms. Overall, the sample of 568 companies consisted of 407 DMNEs and 161 domestic firms. The full sample of 568 companies was used to analyse code adoption rates. Codes of conduct were obtained through a search of the respective company websites, as these are an approved, formal and official expression of corporate policy (Bondy, Matten & Moon, Reference Bondy, Matten and Moon2008). Subsequently, content analysis was performed on all available DMNE codes of conduct. This resulted in a final sample of 179 DMNE codes from 18 countries (see Table 1). Data collection was performed between November 2014 and January 2015.

Table 1:

Code Adoption by Developing Country Firms 2014

Note. Bermuda (9 companies) excluded due to its status as haven; also excluded due to low number of companies in Forbes Global 2000 Index 2014: Bahrain (2 companies), Cayman Islands (1), Egypt (1), Jordan (1), Kazakhstan (3), Kuwait (4), Lebanon (2), Mauritius (1), Morocco (3), Oman (1), Pakistan (2), Peru (2), Puerto Rico (1), Togo (1), Venezuela (2), Vietnam (2).

Data Analyses

Content analysis. Code content was analysed by counting the frequency of an item being mentioned rather than attempting to measure the degree to which it is discussed (Wood, Reference Wood2000). The analysis followed the protocol of a prior study into codes by OECD country firms (Kaptein, Reference Kaptein2004) ^{Footnote 3} but used this in an iterative fashion. The OECD study was taken as a starting point for examining the content of the DMNE codes. However, the content analysis remained open to pick up issues that had not been mentioned in the OECD study. Building on Zenisek’s (Reference Zenisek1979) typology, themes identified in the code content analysis were then grouped into the three dimensions of controversial behaviour, corporate commitments to society and ethical principles.

Once a coding frame had been established, coding was performed on the sample of DMNE codes. Items were coded as ‘1’ if they were referred to at least once in a given code and ‘0’ if no reference was made to the item concerned. Similar words and expressions were considered as long as they carried very close meanings, such as ‘loyalty’, ‘devotion’ and ‘dedication’. Coding was undertaken by two of the three authors. Inter-rater agreement, calculated for a randomly selected sample (accounting for 40% of the overall sample) and expressed as Cohen’s Kappa (Cohen, Reference Cohen1960) was .91.

Quantitative analyses. After we had coded for code adoption, we used logistic regression to explore whether levels of internationalization influence code adoption (i.e. research question 1). For the second research question (i.e. CSR priorities in code content), we first used ANOVA to test if there are differences in content. To validate the differences found in the content analysis, we then conducted exploratory and confirmatory analyses.

Measures

For the dependent variable code adoption, the existence of a code was used to code for adoption in a binary manner (0 = no code, 1 = code). For the independent variable, we coded the level of internationalization as a binary variable (0 = domestic firm, 1 = MNE) based on information from the EBSCO Business and Bureau van Dijk Orbis databases. For the control variables, a recent meta-analysis of the literature on CSR reporting identified three variables, which influence whether firms create such reports: firm size, industry sector and region (Fifka, Reference Fifka2013). Expecting a similar dynamic for codes of conduct, we used these three variables as controls.

For size, we used log of assets (from the Forbes Global 2000 database), which has been used in prior research on international CSR (e.g. Cheng, Ioannou & Serafeim, Reference Cheng, Ioannou and Serafeim2014). For industry sector, we created five dummy variables based on the six main sectors found per Forbes Global 2000 list (i.e. consumer goods, energy & utilities, financials, industrials, information technology & telecommunications and materials; here we used materials as the referent group). For region, we created two dummy variables based on the three main regions in the study (i.e. Latin America and Asia and used other regions as the referent group; China was considered separately given that it constituted by far the largest country subsample).

Code Adoption

Of the 586 developing country companies considered in the sample, 34% made a code of conduct available through their website (see Table 1). For comparison, earlier studies of developed country firms found adoption rates of 92% for the largest 250 firms worldwide (KPMG, 2008) or 98% for the FTSE 350 companies in the UK (Wheldon & Webley, Reference Wheldon and Webley2013), respectively. In our sample, a notable divide emerged between DMNEs and their domestic counterparts: whilst 44% of DMNEs issued a code of conduct, a mere 12 codes could be found among the 161 domestic firms (7.5%).

In addition, other patterns of code adoption can be found. Adoption rates are very unequal within the sample. Adoption among Brazilian, Chilean, Colombian, South African and Thai companies is 80% or higher. In stark contrast, companies from the Middle East and East Asia show very low adoption rates. Only 8 out of 149 Chinese companies (and 7 out of 65 Chinese DMNEs) published a code of conduct. It should be noted, however, that a relatively large number of Chinese companies appear to prefer a different format, such as commenting on core elements of their corporate culture and values, rather than providing a corporate code of conduct. ^{Footnote 4} However, these statements on corporate values are usually light in terms of concrete company commitments and thus do not constitute a functional equivalent of a code of conduct (on the difference between a code of conduct and a value statement see Murphy, Reference Murphy1995).

Confirmatory analysis. We further analysed our results through a logistic regression. Here, we also explored other variables, because sector affiliation and in particular company size may play a role in explaining differences in code adoption between multinationals and their domestic counterparts (a correlation matrix and the results of the logistic regression analysis are presented in Tables 2 and 3). As our results show, levels of internationalization clearly have a positive and significant effect on code adoption. In addition, we can indeed observe clear regional and sector-level differences in code adoption when compared to the reference groups Other Regions and Materials, respectively. Chinese firms stand out in terms of a significantly lower likelihood of issuing a code. Significantly higher levels of code adoption were identified among IT and telecommunication firms, as well as among Latin American firms. However, we found that company size does not have a significant impact on the likelihood of code adoption.

Table 2:

Correlation Matrix (Code Adoption)

n = 179

*p < 0.05; **p < 0.01; ***p < 0.001.

Table 3:

Logistic Regression Model for Adoption of Codes of Conduct as the Outcome Variable

n = 568

R² = 0.26 (Hosmer & Lemeshow), 0.28 (Cox & Snell), 0.38 (Nagelkerke). Model χ²(10) = 541.68, p < 0.001.

*p < 0.05; **p < 0.01; ***p < 0.001. The reference group for Sector is Materials and for Region Other Regions.

Code Content

With regard to code content, our analysis provides support for the assertion that content can indeed be grouped in the three dimensions derived from Zenisek’s (Reference Zenisek1979) work (a summary of code content for the individual country subsamples is provided in Appendices A-D. In the following section, we highlight a few key themes regarding how code content varies in both depth and breadth of coverage. Some of the codes included in the analysis only cover a limited number of key terms, whereas others are substantial documents. For example, the code of Indian telecommunications firm Reliance Communications runs to 56 pages and that of Brazilian steel manufacturer Metalurgica Gerdau has 52 pages. The breadth of code content also reflects a considerable diversity of issues. Saudi Arabian industrial company Savola alerts its employees to the need not to spread gossip; United Arab Emirates NBD Bank prohibits the use of profane language; and Malaysian oil and gas firm Petronas bans cross-dressing.

Despite the variance in depth and breadth, a wide range of recurring themes can be identified across the sample of the codes of conduct (Table 4 summarizes code content along with more descriptive statistics). Not least, our analysis uncovered differences in the degree of coverage among the three dimensions. Codes typically discuss at considerable length, which types of controversial behaviour employees should seek to avoid. On average, 72% of all items in this category are addressed. In contrast, code stipulations regarding the role business should play in society receive the lowest coverage at 18%. Occupying a middle position, 46% of all items in the ethical principles category are addressed.

Table 4:

Code Content 2014 (n = 179)

In addition, our analysis uncovered common themes within each dimension. In the Controversial Behaviour dimension, observing laws, conflicts of interest and confidentiality of information stand out as the three most frequently mentioned items across the sample. Other items that are addressed by at least two-thirds of all codes are bribery, sound accounting, gift giving, protection of company assets, insider information and protection of intellectual property. In the Society dimension, only the two items improving the quality of life for communities and political neutrality emerge as relatively widespread in the sample with coverage of 43% and 39%, respectively. In contrast, the prioritization of items in the Ethical Principles dimension is more diverse. Twenty-one distinct themes were identified, all of which are at least reasonably widespread with coverage levels between 17% and 87%. Fairness/impartiality is the most frequently mentioned item in this category at 87%, followed by honesty/truth at 81% and integrity at 73%.

Beyond these general patterns, there is a considerable amount of country-level variation within the sample (see Appendix D). Whilst controversial behaviour receives the highest coverage in all countries, followed by ethical principles and commitments to society, there are clear differences in the comprehensiveness of codes. Codes by Brazilian and Mexican DMNEs appear most comprehensive and address a wide range of items, in particular in the ethical principles dimension, while Asian (and in particular East Asian) DMNEs provide comparatively little information in their codes. Singaporean DMNEs typically provide no information in the society category; neither do Taiwanese and Saudi Arabian DMNEs. Brazilian DMNE codes prioritize items in the society and ethical principles categories, while Chilean codes emphasize controversial behaviour. Saudi Arabian codes provide relatively little information for any of the three CSR dimensions, whereas DMNEs based in the United Arab Emirates do typically acknowledge a range of items linked to controversial behaviour. Despite these country-level differences within the three categories, a clear pattern can be observed at the level of overall code content: In each of the 18 country subsamples, controversial behaviour is the most frequently addressed category, followed by ethical principles and commitments to society, respectively.

Confirmatory analyses . We then analysed the validity of the findings from the content analysis. We first tested whether differences in dimensions exist among companies. ANOVA results show that mean differences were significantly different for controversial behaviour, F(17, 173) = 5.23, p < .001, commitments to society, F(17, 173) = 4.37, p < .001, and ethical principles, F(17, 173) = 5.26, p < .001. Of the three dimensions, controversial behaviour had the highest coverage (M = .77, SD = .30) compared to commitments to society, (M = .25, SD = .27) and ethical principles (M = .60, SD = .32). This result is thus similar to the findings from the content analysis.

In the content analysis, we adapted an established measure for code content, which had been used for developed countries (Kaptein, Reference Kaptein2004), but not for developing ones. Based on the work of Zenisek (Reference Zenisek1979), we then categorized the items into three categories (i.e. controversial behaviour, corporate commitments to society and ethical principles). In order to test whether a three-factor structure is appropriate, we performed an exploratory factor analysis (EFA) using Principal Axis Factoring and orthogonal Varimax rotation. In order to avoid using the same sample for both the exploratory and confirmatory factor analyses, we used a split-sample design (e.g. Koh, Steers & Terborg, Reference Koh, Steers and Terborg1995) in which the sample was randomly split into two. The split-half sample for the exploratory factor analysis consisted of codes of conduct by 94 companies from 18 countries. The split-half sample for the confirmatory factor analysis included 97 codes from 18 countries.

The Bartlett test of sphericity was significant, χ² (105) = 304.34, p < .001. In addition, Kaiser-Meyer-Olkin as well as anti-correlation matrix values for individual items were above the recommended cut-off value of 0.50 (Field, Reference Field2000). Moreover, the Kaiser-Meyer-Olkin measure of sampling adequacy was 0.72 (p < .001). Therefore the data showed that there was a patterned relationship between the items; or in other words, the data was sufficient for EFA (Field, Reference Field2000). Using an eigenvalue cut-off of 1.0, there were three factors that explain a cumulative variance of 47%. The scree plot confirmed that the three factors could be retained. We then removed items that did not fit the factors well, such as those that loaded on the factor below a cut-off value of 0.4 (Stevens, Reference Stevens1992).

Next, we conducted a confirmatory factor analysis (CFA) on the second sample that resulted from randomly splitting the original sample (n = 97). Despite the low statistical power, the model showed good fit according to recommended values (Hu & Bentler, Reference Hu and Bentler1999). The root-mean-square error of approximation (RMSEA) was 0.046 with 90% confidence intervals of 0.000 and 0.075 (Hu & Bentler, Reference Hu and Bentler1999). The standardized root mean square residual (SRMR) for the model was 0.074. The comparative fit index (CFI) for the model was 0.94 and the Tucker and Lewis index (TLI) was 0.93. Difference tests showed that a three-factor model was better than a two-factor model (χ² difference = 110.89, df = 44, p < .001) and a one-factor model (χ² difference = 293.46, df = 89, p < .001).

In summary, Study 1 uncovered company-level patterns that exist within the sample of developing country codes, not least that DMNEs are much more likely to adopt a code than their domestic counterparts.

Political System

According to Whitley (1999: 48), countries differ in the extent to which governments “directly or indirectly regulate market boundaries, entry and exit, as well as set constraints on the activities of economic actors”, whether through pursuing active industrial policies, supporting collective intermediate organizations or the regulation of markets. While some states, such as France, Japan or more recently China (Nee, Opper & Wong, Reference Nee, Opper and Wong2007), have at various times pursued an active industrial policy, many developing countries suffer from inefficient government organizations, high levels of corruption or even civil unrest (Hamann, Kapelus, Sonnenberg, Mackenzie & Hollesen, Reference Hamann, Kapelus, Sonnenberg, Mackenzie and Hollesen2005). Therefore, we would expect the content of codes of conduct by DMNEs from different countries to be shaped by key factors of the political system, like the quality of political institutions. Following the ‘substitute view’ of CSR as explained above (Hiss, Reference Hiss2009; Jackson & Apostolakou, Reference Jackson and Apostolakou2010), we expect companies based in countries characterized by weak political institutions to communicate a wider set of responsibilities in their codes of conduct.

Financial System

Following Whitley (1999: 49), the financial system “deals with the processes by which capital is made available and priced. In particular, is it allocated by capital markets through competition […] or is it provided by some set of intermediaries that deal directly with firms and become locked into their particular success?” In many developing countries stock markets and institutional investors play a much smaller role than in the Anglo-American system and many emerging economies have also been more volatile, which leaves firms with fewer financing options (Céspedes, González & Molina, Reference Céspedes, González and Molina2010). For example, Chilean firms have been found to often operate as business groups with common family ownership among all the group’s firms because this organizational form overcomes the lack of financing through capital markets (Silva, Majluf & Paredes, Reference Silva, Majluf and Paredes2006). These differences between countries again can be expected to have repercussions for the CSR-related commitments firms make. In line with the ‘substitute view’ of CSR (Hiss, Reference Hiss2009; Jackson & Apostolakou, Reference Jackson and Apostolakou2010), we would expect weaker financial market-related institutional structures to trigger higher levels of CSR engagement by firms from these countries. Companies are therefore likely to communicate a wider set of responsibilities in weaker macroeconomic environments.

Labour System

Whitley (1999: 50) furthermore argues that NBSs differ in terms of “the system that develops and certifies competences and skills: the education and training system” as well as “the institutions that control the terms on which the owners of those skills sell them in labour markets and how those markets are organized”, i.e. the labour system. According to the World Economic Forum (2014), efficient labour markets promote meritocracy in the workplace and “allow for wage fluctuations without much social disruption” (World Economic Forum, 2014: 7). In this respect, key differences between developed and developing countries can again be observed, although there are of course huge differences within either group, too. Many developing countries face challenges in terms of labour market efficiency. Earnings levels are often low despite long working hours, income is often more uncertain and there is a greater reliance on non-wage labour, particularly in agriculture. Moreover, in many countries women are particularly disadvantaged in getting access to labour markets (Fields, Reference Fields2011). These differences in labour market efficiency are likely to influence the content of a firm’s code of conduct. Following the ‘substitute view’ of CSR again, we expect code content to be more comprehensive in countries where labour market efficiency is low.

Sample

The same sample of 179 DMNE codes from 18 countries used in Study 1 was also used for Study 2 (see Table 1). However, the software that we used for multilevel modelling (HLM Version 7; Raudenbush, Bryk, Cheong, Congdon & du Toit, Reference Raudenbush, Bryk, Cheong, Congdon and du Toit2011) deletes groups (i.e. countries in our study) if there is missing data at the group level (i.e. in our study independent variables at the country level). This process resulted in a final sample for Study 2 of 164 DMNE codes from 15 countries.

Measures

Dependent variable: coverage in codes of conduct. We used the measure that resulted from the confirmatory factor analysis in Study 1. The final measure consists of 15 items, with 5 items for each of the three factors (i.e. controversial behaviour, commitments to society, ethical principles). As explained previously, they were coded as ‘1’ if they were referred to at least once in a given code and ‘0’ if no reference was made to the item concerned. We then used the mean of the items, resulting in a variable that measures the amount of coverage in codes of conduct.

Independent variables. Building on previous studies (Ioannou & Serafeim, Reference Ioannou and Serafeim2012; Whitley, Reference Whitley1999), a number of additional independent variables were used to identify relationships between the elements of a National Business System and code content. For Hypotheses 1, 2 and 3, we used data from the 2014-2015 Global Competitiveness Report (World Economic Forum, 2014), which has been and used in prior research (e.g. Shaner & Maznevski, Reference Shaner and Maznevski2011; Wan & Hoskisson, Reference Wan and Hoskisson2003). Specifically, for the political system (Hypothesis 1), we used quality of political institutions (World Economic Forum, 2014). For the financial system (Hypothesis 2), we used quality of the macroeconomic environment (World Economic Forum, 2014). For the labour system (Hypothesis 3), we used labour market efficiency (World Economic Forum, 2014).

Control variables. We used the same control variables as in Study 1: log assets of size and dummy variables for industry sector and region. In order to build on Study 1, we also included a control variable for the level of internationalization, which explored the influence of internationalization on code coverage. For levels of internationalization, we used foreign sales divided by total sales (from the Bureau van Dijk Orbis database), with the purpose to partial out international effects so that country effects could be more accurately analysed.

Data Analyses

Study 2 focuses on the effects of NBS elements on companies, which are headquartered in these countries. In order to test our hypotheses and properly capture between-country differences, we employed hierarchical linear modelling (HLM Version 7; Raudenbush et al., Reference Raudenbush, Bryk, Cheong, Congdon and du Toit2011). As recommended by Bryk and Raudenbush (Reference Bryk and Raudenbush1992), our HLM model consisted of two levels. Company-level data was modelled at level 1 and country characteristics at level 2. Following recommended procedures for centring in order to facilitate interpretation and also to reduce multicollinearity, level-1 variables (i.e. size, level of internationalization) were group-centred except for dummy-coded variables (i.e. industry, region), while level-2 variables were grand-mean centred (Hofmann & Gavin, Reference Hofmann and Gavin1998). Due to high correlations between variables, we calculated variance inflation factors (VIF). The highest VIF was 1.76, which is well below the recommended cut-off value of 10, suggesting that multicollinearity was not an issue (Ryan, Reference Ryan1997). We used restricted maximum likelihood estimation. While estimation techniques will provide similar results for larger samples, for smaller samples such as ours it is important to choose the more reliable approach (Raudenbush & Bryk, Reference Raudenbush and Bryk2002). Therefore we chose restricted maximum likelihood, because it reduces bias in the estimates of the variances (Snijders & Bosker, Reference Snijders and Bosker2012). However, we also tested our data with full maximum likelihood, finding very similar results.

Hypotheses Testing

Table 5 shows descriptives and a matrix of correlations of the variables employed in this study, while Table 6 displays the HLM results. We first tested a Null model with only the dependent variable in order to examine whether a multilevel model is necessary. The overall intra-class correlation (ICC) is 0.33; in other words, 33% of the variance in code coverage resided between countries. Then, in Model 1 of the HLM analysis, we included only the control variables. Finally, in Model 2, we added the independent variables. Dummy codes for industry and region are not displayed in Table 6 for purposes of brevity; however we included them in the analysis and none of the dummy variables were significant in either Model 1 or 2. Because HLM does not provide effect sizes when using restricted maximum likelihood, we followed procedures recommended by Hofmann (Reference Hofmann1997) to calculate effect size. Model 1 explained 34% of the variance in the dependent variable (R Footnote ² = 0.34), while Model 2 showed an improvement with 83% of the variance explained (R Footnote ² = 0.83).

Table 5:

Correlation Matrix (Code Content)

n = 179

*p < 0.05; **p < 0.01; ***p < 0.001.

Table 6:

Hierarchical Linear Models for Coverage of Codes of Conduct as the Outcome Variable

n = 164 companies at Level 1; n = 15 countries at Level 2; SE = Standard errors

*p < 0.05; **p < 0.01; ***p < 0.001.

(a) The analyses included dummy variables for sector and region as control variables, but are not shown in the Table above.

Hypothesis 1 was supported as quality of political institutions was significantly and negatively related to code coverage (γ = -0.46, p < .01). Hypothesis 2 was supported as quality of the macroeconomic environment was significantly and negatively related to code coverage (γ = -0.17, p < .01). Hypothesis 3 was not supported as labour market efficiency was significantly but positively related to code coverage (γ = 0.55, p < .01), whereas we had hypothesised a negative relationship.

Post Hoc Analysis

We also analysed the full sample that included both multinationals and domestic companies (n = 191) and results for all the hypotheses in Study 2 were similar. Quality of political institutions was significantly and negatively related to code coverage (γ = -0.42, p < .01). Quality of the macroeconomic environment was significantly and negatively related to code coverage (γ = -0.18, p < .01). Labour market efficiency was significantly and positively related to code coverage (γ = 0.60, p < .01).