Who sets the CSR agenda? : the interplay between the news media, organizational public relations, and the public

Who sets the CSR agenda?

The interplay between the news media, organizational

public relations, and the public

Linda van den Heijkant 10210997

Master’s Thesis

Research Master’s programme Communication Science Graduate School of Communication

Supervisor: Prof. dr. Rens Vliegenthart Date of Completion: February 3, 2017

Abstract

Despite the widely accepted idea that organizations have enormous responsibilities toward society, there is an ongoing discussion about the concept and practice of Corporate Social Responsibility (CSR). The news media, organizational public relations (PR), and the public can be considered as key domains in which this discussion about CSR takes place. However, little is known regarding the communication about CSR and the interplay between the different domains. Therefore, this study applies a longitudinal design that allows examining the multidirectional causal relationships between the news media, organizational PR, and the public. Drawing on, among others, agenda-setting and framing theory, the purpose of the study was to investigate whether the domains affect each other in terms of salience and implicit frames of CSR. To empirically investigate the salience and implicit frames of CSR, an automated content analysis of newspaper articles and organizational press releases is employed. Additionally, Google Trends data was examined in order to determine the salience and implicit frames of CSR among the public. Monthly level vector autoregression (VAR) analyses show, with regard to salience of CSR, that the news media negatively affect organizational PR and the public, whereas the public negatively affects the news media. In addition, organizational PR positively affects the public, while evidence for the causal effects of organizational PR on the news media and the public on organizational PR is not found. With regard to the implicit frames of CSR, different effects across frames are found. Overall, results point in the direction of positive effects of the news media on the public and

organizational PR on the news media.

Keywords: corporate social responsibility, agenda-setting, agenda-building, framing,

frame-building, automated content analysis, implicit frames, domains, media, organizational PR, the public.

Introduction

Throughout the years, a variety of perspectives on the concept and practice of Corporate Social Responsibility (CSR) have emerged (Amaladoss & Manohar, 2013). These

perspectives have one particular aspect in common: the acknowledgment that organizations have enormous responsibilities toward society that go beyond their own concerns and legal requirements (Amaladoss & Manohar, 2013; Podnar, 2008). However, due to changing norms, values, and expectations over the years, there is an ongoing discussion about what CSR should entail and in which way it should be communicated (Tam, 2015). The news media are acknowledged to be one of the domains in which this debate on CSR occurs (Tam, 2015). Besides the news media, organizational public relations (PR) is a domain with a crucial role in the discussion about CSR (Du, Bhattacharya, & Sen, 2010). The socially responsible business activities are communicated through public relations tools such as annual reports and press releases. Another significant domain in which the debate on CSR takes place is the public realm (Burchell & Cook, 2006). The public is participating in both the consumption and production of CSR communication (Capriotti, 2011).

Several studies have focused on the key role played by media, organizations, and the public in the communication about CSR (e.g. Carroll, C., 2011; O’Conner & Shumate, 2010; Lee & Carroll, C., 2011). However, most studies on CSR communication have examined the communication within merely one domain (Tam, 2015). Therefore, little is known regarding the communication about CSR in the interplay between the different domains. Moreover, existing research in the field of CSR communication is mainly cross-sectional (Aguinis & Glavas, 2012). A cross-sectional design does not provide the opportunity to investigate links between different domains, while this is interesting and relevant to study. In addition,

longitudinal studies of the relationship between information subsidies (e.g. press releases) and the news media are scarce (Boumans, Vliegenthart, & Boomgaarden, 2016). The current

study will fill those gaps by applying a longitudinal design that allows to investigate the over-time dynamics of communication about CSR within the three domains and the reciprocal relationships between these domains.

Drawing on, among others, agenda-setting and framing theory, the aim of the study is to investigate whether the domains affect each other in terms of salience and use of implicit frames of CSR. Within this interplay, the focus will be on three particular multidirectional relationships: (1) the relationship between the news media and the public, (2) the relationship between the news media and organizational PR, and (3) the relationship between

organizational PR and the public. These relationships will be examined over a period of five years, ranging from January 2011 to December 2015. The longitudinal design of this study is beneficial since it provides the possibility to reflect a much more dynamic perspective of the evolvement of CSR as an issue (Lee & Carroll, C., 2011). Moreover, a longitudinal design explicates temporal ordering, getting one step closer to disentangling the causal direction of relationships between the domains. The overall research question is as follows: To what extent

does the communication about CSR in the domains news media, organizational PR, and the public affect each other in terms of salience and implicit frames?

In order to obtain an answer to the research question, an automated content analysis of newspaper articles and organizational press releases will be executed. In addition, Google Trends data will be examined in order to determine the public attention to CSR. The study will theoretically contribute to the field of CSR communication by examining the mutual relationships between the news media, organizational PR, and the public. This study seeks to get one step closer to distinguishing the causal direction of relationships between these domains. Besides the theoretical contribution, this study attempts to methodically enrich the field of CSR communication by employing an automated content analysis of implicit frames. Gaining more insights into the CSR communication within the interplay of the domains will

be valuable for public relations practitioners, media practitioners and the public. The findings of this study can offer actors within the domains a better understanding of their role in CSR communication and how they, as a domain, affect other domains in this CSR communication.

Theoretical Background Dimensions of CSR

Although a universal all-embracing definition of CSR is lacking (Tam, 2015; Van Marrewijk, 2003), CSR can be defined by means of various responsibilities of organizations toward society (Carroll, A, 1979). These responsibilities of organizations toward society can be distinguished into four dimensions: economic, legal, ethical, and philanthropic (Carroll, A., 1979; Lee & Carroll, C., 2011). First, the economic responsibility within society demands organizations to make profit and to offer goods and/or services that society needs (Carroll, A., 1979; Lee & Carroll, C., 2011). This type of responsibility is characterized by creating value for shareholders and stakeholders. Creating jobs and fair pay for employees and facilitate a return on investment to shareholders are examples of this specific type of responsibility (Jamali & Mirshak, 2007). Second, the legal responsibility of an organization entails the expectation of society that an organization is operating within the framework of legal requirements (Carroll, A., 1979; Lee & Carroll, C., 2011). This part of CSR focuses on the responsibility of organizations to play by the rules of the law (Jamali & Mirshak, 2007). An organization that creates a good working environment for its employees can be seen as an organization that is meeting its legal responsibility (Lee & Carroll, C., 2011). Third, society expects organizations to behave in a way that is morally right, which is referred to as the ethical responsibility of an organization (Carroll, A., 1979; Lee & Carroll, C., 2011). This entails business activities that are not covered by the law, but are, nevertheless, expected by society (Jamali & Mirshak, 2007). Avoiding harm for the environment is an example of this

type of responsibility. Finally, philanthropic responsibility requires organizations to respond to the wish of society to be actively involved in the enhancement of society. Corporate volunteerism and donating to charity are examples of this type of responsibility (Lee & Carroll, C., 2011).

CSR communication by the news media, organizational PR, and the public

CSR is an ambiguous issue to which actors can ascribe various meanings for different intentions (Tam, 2015). In this regard, the communication about CSR can differ across multiple actors and domains. There are three domains that can be considered as key arenas in which the communication about CSR takes place. First, the news media are an important domain involved in the discourse on CSR (Du et al., 2010; Tam, 2015). Since the 1970s, the news media gained interest in and started to communicate about CSR (Buhr & Grafström, 2006; Hamilton, 2003). The news media can be viewed as a significant arena for the public sphere (Carroll, C., 2011). Second, organizational PR is a prominent domain involved in the communication about the concept and practices of CSR (Du et al., 2010). CSR occupies a high position on the corporate agenda of organizations worldwide (Du et al., 2010). Businesses are increasingly devoting significant amounts of resources to various social initiatives (Du et al., 2010). Public relations tools such as annual CSR reports, press releases, or a section on the corporate website are examples of ways in which organizations

communicate about CSR (Du et al., 2010; Perks, Farache, Shukla, & Berry, 2013). Finally, the public, i.e. civil society, is a significant domain in which the discourse on CSR is shaped (Burchell & Cook, 2006). Due to the rise of the Internet and social media, people can initiate or participate in a discussion in a relatively simple and rapid manner (Neuman, Guggenheim, Jang, & Bae, 2014). As a consequence, besides participating in the consumption of CSR communication, the public started to participate in the production of CSR communication as

well (Capriotti, 2011).

Communicators within these domains attempt to gain access to and influence the debate on CSR (Tam, 2015). Each domain consists of numerous communicators and has its own dialogue on CSR. Notwithstanding, these dialogues of the various domains are linked to each other (Tam, 2015). New communications about CSR are linked to communications about CSR that have been previously published, as well as to communications that are simultaneously or subsequently published (Dempsey, 2011). This implies that any communication about CSR might be influenced by CSR discussions in the past, and can possibly influence the dialogue on CSR in the future. Therefore, rather than considering the CSR communication of the different domains as isolated from each other (Tam, 2015), the key interest of this study is to regard these communications as being linked to each other.

Figure 1 visualizes the interplay between the domains news media, organizational PR, and the public. This study will apply an open approach in which every relationship will be tested in both directions since knowledge about these relationships in the field of CSR

communication is limited. Thus, despite that some relationships are expected to be less likely than others, all relationships will be examined in both directions for reasons of consistency and completeness. In the following sections, each relationship within this interplay will be outlined in both directions in terms of salience and implicit frames of CSR.

News media and the public Salience of CSR

The news media are active actors in, and a domain for, shaping information (Deephouse, 2000; Fombrun & Shanley, 1990). Therefore, this domain has an important influence on the focus of public attention (López-Escobar, Llamas, & McCombs, 1998). Traditionally, agenda-setting research has investigated the transfer of salience of issues from

Figure 1. Interplay between the news media, organizational PR, and the public

Note. The solid black arrows represent the relationships with a strong theoretical basis for which hypotheses are formulated; the open white arrows represent the relationships with a less strong theoretical basis for which research questions are formulated.

the media to the public (McCombs & Shaw, 1972). The core proposition of agenda-setting theory is that the emphasis on issues in the news media leads to an increase in the concern with those issues among the public (e.g. Carroll, C., & McCombs, 2003; Kiousis &

McCombs, 2004). This implies that by deciding which issues will be highlighted, the news media determine what the public thinks about (e.g. McCombs, 2005; McCombs & Shaw, 1972; Tam, 2015). The transfer of issue salience from the media to the public is referred to as first level agenda-setting (e.g. McCombs, 2005; Kiousis, Mitrook, Wu, & Seltzer, 2006). An example of an issue that can be transferred from the media to the public is CSR. According to first level agenda-setting, prominence of CSR in the news media will lead to an increase in the public attention to CSR. Hence, the following expectation is formulated:

H1a. The salience of CSR in the news media positively affects the salience of CSR among the public.

Besides traditional agenda-setting, which is widely accepted and the most often investigated direction, another path of agenda-setting is the bottom-up approach (McCombs, 1997). The bottom-up approach is also referred to as reversed agenda-setting (McCombs,

2004). Reversed agenda-setting suggests, contrary to traditional agenda-setting, that the public agenda precedes and influences the media agenda (McCombs, 2004; Neuman et al., 2014). This implies that the public interest in certain issues leads to an increase in the

attention for those issues in media coverage. Although the concept of reversed agenda-setting has been known and investigated earlier (e.g. Kleinnijenhuis & Rietberg, 1995), the rise of the Internet plays a major role in the principle of reversed agenda-setting (Guo & McCombs, 2015) and motivated researchers to focus on this subject (e.g. Kim & Lee, 2007). Due to the rise of the Internet, people are not merely receivers of communication, but also active

participants and initiators in discussions (Jiang, 2014; Neuman et al., 2014). The Internet has provided the public a platform where they can engage in mass-to-mass communication (Castells, 2007). This is denoted as mass self-communication and enabled the public to generate and distribute content to a potentially global audience (Castells, 2007). Thus, by using the Internet and new communication technologies such as social media, people can communicate and propose issues on the agenda, which will be taken over by the news media (Chaffee & Metzger, 2001; Jiang, 2014). The concept of reversed agenda-setting has,

however, not been studied in the field of CSR communication up to now. Moreover, in general, there is not much scientific evidence for the notion of reversed agenda-setting. Despite the fact that previous research provide some indications, it is uncertain whether the concept of reversed agenda-setting can be applied to the context of CSR communication. Therefore, rather than proposing a hypothesis, this relationship will be explored by means of the following research question:

RQ1a. How does the salience of CSR among the public affects the salience of CSR in the news media?

Implicit frames of CSR

Besides determining the salience of CSR among the public, the media may also shape public opinion about CSR by framing issues in a specific way (De Vreese, 2005). Framing can be defined as follows: “to select some aspects of a perceived reality and make them more salient in a communicating text, in such a way as to promote a particular problem definition, causal interpretation, moral evaluation, and/or treatment recommendation” (Entman, 1993, p. 52). Framing thus entails the selection and salience of some aspects of an issue (Entman, 1993). Through accentuating certain aspects of an issue above others, a frame provides a manner to interpret an issue (De Vreese, 2005).

Given that framing entails the selection and salience of some aspects of an issue (De Vreese, 2005; De Vreese, 2012), framing can be used to influence the public how to think about an issue (Hellsten, Dawson, & Leydesdorff, 2010; McCombs, 2005). Hence, framing is comparable to second level agenda-setting, which is concerned with the salience of attributes of specific issues (Carroll, C., & McCombs, 2003; McCombs, 2005). This implies that the media can influence the public by deciding which attributes of issues are emphasized and which attributes are completely ignored or mentioned briefly (McCombs, 2005; Kiousis et al., 2006). In a similar vein, it can be argued that the news media play a role in influencing

people’s opinions about CSR by making certain aspects of CSR more salient through frames. Thus, the way CSR is addressed in the news media affects the way it is interpreted by the public, since the news media put information into a context and determine the frames of reference in which people evaluate the information, understand meanings and even take actions when it is appropriate (Hallahan, 1999).

A frame can be either explicit or implicit (Hellsten et al., 2010; Van der Meer, Verhoeven, Beentjes, & Vliegenthart, 2014). Since most of the frames that are applied in public debates are not explicit (Hellsten et al., 2010), the current study will focus on implicit

frames. The distinction between explicit and implicit frames will be discussed in the method section. In line with framing theory and second level agenda-setting, the following hypothesis is formulated:

H1b. The implicit frames used by the news media regarding CSR positively affect the use of these implicit frames among the public.

As mentioned earlier, mass self-communication has a potential reach of a global audience through the Internet and includes self-generated content by the public (Castells, 2007). In line with the idea of mass self-communication, citizen journalism focuses on the power of people to engage in journalistic practices and participate in the news process (Goode, 2009). Hence, rather than staying ‘nobodies’, the public gained attention online and became ‘somebodies’ (Booth & Matic, 2011). Consequently, the framing power of the public as a domain has increased. The news media observe, and possibly use, the stories produced by the public (Lariscy, Avery, Sweetser, & Howes, 2009). In this regard, it can be argued that the public’s attention for particular attributes of specific issues, i.e. the focus on certain implicit frames, influences the emphasis on these particular attributes in the media. However, there is no existing research in the context of CSR communication that can be used to support this reasoning. Hence, this study will explore the relationship and therefore the subsequent research question is formulated:

RQ1b. How does the use of implicit frames of CSR among the public affect the use of these implicit frames by the news media?

News media and organizational PR Salience of CSR

the salience of certain issues in the domains of news media and organizational PR. Agenda-building is a process by which needs and demands of groups in society are translated into items competing to get attention of public officials (Cobb, Ross, & Ross, 1976). Agenda-building can, thus, be outlined as the mutual connections among various stakeholder groups in the process of formation, transfer, and exchange of salience of issues (Kiousis et al., 2015). These stakeholder groups comprise policymakers, interest groups, activists, and so forth. Particularly relevant for this study, these groups include news media and businesses as well (Kiousis et al., 2015). A principal strategy for businesses to generate news attention in the agenda-building process is by means of the use of information subsidies (Kiousis et al., 2015). Information subsidies such as organizational press releases are an important source for the media (Davis, 2000; Deephouse, 2000; Kiousis et al., 2006). Organizations can use their press releases as a means of disseminating information to the public via the news media (Carroll, C, 2011; Tam, 2015; VanSlyke Turk, 1985). Due to the fact that news media are communicators that cannot be entirely controlled by organizations, the news media are more credible in communicating CSR information compared to organizations themselves (Du et al., 2010). The less controllable the communicated information is for an organization, the more credible the information is perceived by the public (Du et al., 2010). Therefore, organizations attempt to encourage the media to take over the information they spread through their communications such as press releases (Carroll, C., & McCombs, 2003; Du et al., 2010). When organizational PR succeed in convincing the news media to take over the information in press releases, organizational PR is influencing the media agenda (and the public agenda) (Sallot & Johnson 2006). In this regard, organizational PR can influence the salience of CSR among the news media.

The relationship between organizational PR and the news media is, however, not unidirectional (Kiousis, Popescu, & Mitrook, 2007). The news media and organizational PR

exchange information and impact each other in the process of newsgathering (Kiousis et al., 2007; Sallot & Johnson, 2006). Hence, the other way around, the news media are an

important source of information for organizational PR as well (Sallot & Johnson, 2006). By exchanging information, the news media can, for instance, be an important source on which issues are important and which are not. Moreover, according to Hollanders & Vliegenthart (2008), news becomes news because it was news yesterday. This is based upon the idea that issues come and go in cycles (Downs, 1972) and hypes (Vasterman, 2005). In order for a press release to be adopted by the media, organizational PR might use the news media to check which issues received attention. Consequently, organizational PR will present these issues in their press releases to increase the likelihood that the news media will adopt their coverage. In this regard, the news media may influence organizational PR as well. Therefore, the following hypotheses are formulated:

H2a. The salience of CSR in the news media positively affects the salience of CSR in organizational PR.

H2b. The salience of CSR in organizational PR positively affects the salience of CSR in the news media.

Implicit frames of CSR

Framing implies, as mentioned earlier, the selection and salience of certain aspects in a news report (Entman, 1993). Framing involves thus the inclusion and exclusion of certain aspects as well as the emphasis on these aspects (Hallahan, 1999). But how do particular frames emerge? Since communication is a dynamic process rather than a stable process, it involves frame-building (De Vreese, 2005). Frame-building concerns the processes in which frames emerge and gain prominence (Scheufele, 1999; Vliegenthart & Van Zoonen, 2011). The frame-building process occurs in an ongoing interaction between journalists, (political)

elites, and societal actors such as social movements, interest groups and businesses, with the frames that occur in the communication text as an outcome of this process (De Vreese, 2005). The way CSR is communicated within the news media is thus a dynamic interaction between several factors. Factors that influence this process are both factors internal as well as external to journalism (De Vreese, 2005). According to Shoemaker and Reese (1996), an external domain that can exert influence on the framing of the news is for example organizational PR. Organizational PR are highlighting particular aspects of an issue within their press releases to influence certain stakeholders, such as the news media (Gilpin, 2008). Since the news media are influenced by information subsidies such as organizational press releases, organizations can influence the use of implicit frames among the news media. Thus, frames suggested by organizational PR about a particular issue (e.g. CSR) are utilized and integrated in the communication about that issue (e.g. CSR) in the news media (Scheufele, 1999). Vice versa, the news media can also influence the implicit frames used in press releases. As mentioned earlier, news becomes news because it was news yesterday (Hollanders & Vliegenthart, 2008). Therefore, in order for a press release to be adopted by the media, organizational PR should frame information in their press releases in such a way that it fits previously presented information in the media. In this way, it can be argued that the news media are influencing the use of implicit frames among organizational PR. This results in the following hypotheses:

H2c. The implicit frames used by the news media regarding CSR positively affect the use of these implicit frames by organizational PR.

H2d. The implicit frames used by organizational PR regarding CSR positively affect the use of these implicit frames by the news media.

Organizational PR and the public Salience and implicit frames of CSR

Unfortunately, in scholarly research, little is known about how organizational PR contributes to the salience of issues and salience of attributes of issues among the public (Kiousis et al., 2006). In the field of political communication, Kiousis et al. (2006) gained evidence for the effect of the salience of issues and candidate images in campaign news releases on the public opinion regarding these issues and candidates (Kiousis et al., 2007). However, in general and especially in the field of CSR communication, studies of the relationship between organizational PR and the public are very scarce or even nonexistent. Rather than focusing on the direct relationship between organizational PR and the public, research to date focused mostly on the indirect relationship between organizational PR and the public through the news media, as outlined above. This indirect relationship between organizational PR and the public is, however, already captured within the interplay in this study. Since previous research is limited to nonexistent, the direct relationship between organizational PR and the public in terms of CSR salience and the use of implicit frames will be explored in this study. Hence, the following research questions were formulated:

RQ2a. How does the salience of CSR in organizational PR affects the salience of CSR among the public and vice versa?

RQ2b. How does the use of implicit frames of CSR in organizational PR affect the use of these implicit frames among the public and vice versa?

Within this theoretical framework, some relationships are strongly theorized while other relationships have a less strong theoretical basis. However, since there is little knowledge about the interplay of the news media, organizational PR, and the public in the field of CSR communication, an open approach will be applied. Therefore, all relationships visualized in Figure 1 will be examined in both directions. In this way, the interaction between the domains news media, organizational PR, and the public will be fully tested.

Methods

Data collection part I: press releases and newspaper articles

An automated content analysis is used to empirically investigate the salience and implicit framing of CSR in the news media and organizational communication. With this automated content analysis, newspaper articles and organizational press releases were analyzed. With regard to the organizational press releases, two large scale Dutch organizations of every major sector – heavy industry, retail, and service sector – were

selected. By including different types of organizations in the sample, a general impression of Dutch organizations was gained. Based on selection criteria1_{, Shell and AkzoNobel were}

selected for the heavy industry sector, KPN and Delta Lloyd were selected for the service sector, while Jumbo and FrieslandCampina were selected for the retail sector. This sample can be considered to be representative of all large scale Dutch organizations, and this enhances the generalizability of findings. The press releases of the years 2011 to 2015 were retrieved from the corporate websites of the organizations and systematically explored by the researcher in order to determine which press releases contained relevant information in the context of CSR. The criteria that were used to decide whether a press release included

relevant information can be found in Appendix A. This selection procedure resulted in nearly five hundred relevant press releases (N = 494), Shell (n = 113), AkzoNobel (n = 90), KPN (n = 79), Delta Lloyd (n = 49), Jumbo (n = 40), and FrieslandCampina (n = 123). To check the reliability of the selection procedure, a second person repeated the selection process for a sample of forty press releases. The sample consisted of press releases of the different years and organizations, with an equal distribution of relevant and irrelevant press releases. Since the percentage agreement (100 percent) and the Cohen’s Kappa and Krippendorff’s Alpha (1.00) indicated a perfect agreement, it can be assumed that the selection procedure of the press releases is highly reliable.

Besides organizational press releases, this study relies on newspaper articles. Newspapers were the most appropriate type of news media for this study, since newspaper articles are available as text documents and easily accessible through the online database LexisNexis (Van der Meer, 2016). Moreover, newspapers are still widely read in the

Netherlands (Bakker, 2013) and considered to be important agenda-setters (De Feijter, 2007). The newspaper articles were obtained by means of the online database LexisNexis. In order to obtain as many relevant newspaper articles as possible while avoiding irrelevant articles, a variety of search strings was tested, adjusted and fine-tuned. Eventually, the most appropriate search string was used to collect the newspaper articles related to CSR2_{. The time period}

ranged from January 2011 to December 2015 and the sources of the newspaper articles were restricted to the six most-read Dutch national newspapers, which are the Telegraaf, Metro, Algemeen Dagblad, Volkskrant, NRC Handelsblad, and Trouw. This search string provided in total 1286 newspaper articles, 2011 (n = 318), 2012 (n = 306), 2013 (n = 282), 2014 (n = 207), and 2015 (n = 173).

Automated content analysis

A frame can be, as mentioned before, either explicit or implicit (Hellsten et al., 2010; Van der Meer et al., 2014). Explicit frames are recognizable in the words that are selected in a text, while implicit frames are concerned with latent patterns of words that co-occur in a text (Van der Meer et al., 2014). With regard to implicit frames, meanings of words are thus constructed in the context in which they are utilized (Hellsten et al., 2010). This study focused on implicit frames rather than explicit frames, since most of the frames that are applied in the public sphere are not explicit (Hellsten et al., 2010). This means that the majority of frames used in public discussions are embedded in latent dimensions of communication rather than clear-cut presented perspectives (Hellsten et al., 2010). Different from explicit frames,

implicit frames cannot be easily detected with the aid of a manual content analysis (Hellsten et al., 2010). Therefore, an unsupervised automated content analysis of newspaper articles and organizational press releases was executed. Unsupervised methods refrain from analyzing the newspaper articles and press releases on the basis of a priori defined frames (Van der Meer, 2016). Rather, the theoretical framework is tested by inductively identifying the word

clusters, i.e. implicit frames, in the text (Van der Meer, 2016). This method makes it possible to investigate frame development over time and to compare frames between several domains (Van der Meer, 2016).

Prior to conducting the automated content analysis, some pre-processing steps needed to be followed. First, the software dtSearch is used to split the newspaper articles and press releases into separate documents. This resulted in separate text files for each article.

Additionally, the fields ‘organization’ or ‘newspaper’, ‘headline’, and ‘date’ were extracted from the text documents based on specific markers in the text. After these pre-processing steps, the automated content analysis began by identifying the most frequently used words in the texts about CSR. Frequency count of words is a straightforward procedure to determine the relative attention for an issue such as CSR (Krippendorff, 2004). In addition, frequency count is advantageous in determining whether a phenomenon such as CSR communication varies across time and sources (Sonpar & Golden-Biddle, 2008). A list of all words and corresponding frequencies was established through the software dtSearch for both the press releases and newspaper articles. From each list, three coders indicated the 250 most

frequently used words that were related to CSR or could be relevant in the context of CSR, while omitting stop words, numbers and remaining irrelevant words. The words that were indicated by at least two of the three coders were selected and this resulted in a list of 361 words (for a comparable approach, see McLaren, Boomgaarden & Vliegenthart, 2017). Next, the 361 words were included in a dictionary constructed in the content analysis program

Yoshikoder. This dictionary has been inserted in the software program JFreq, and

subsequently all newspaper articles and press releases were content analyzed based on the dictionary. This resulted in two word matrices, which were converted into a SPSS file, in which each of the 361 words constituted a variable, each text was a case, and the scores represented the number of occurrences of each relevant word per text.

Next, the salience of CSR within the domains of news media and organizational PR was derived. The salience of CSR within organizations was operationalized as the number of published press releases in the period of January 2011 to December 2015, whereas the salience of CSR within the news media was operationalized as the number of published newspaper articles in the same period. Subsequently, the presence of the CSR issue was aggregated to monthly levels, resulting in variables indicating the monthly salience of CSR in press releases as well as newspaper articles.

With regard to the implicit frames, a principal-component factor analysis with varimax rotation was conducted to identify word clusters. Words that load high on the same factor co-occur often in the same context and, therefore, can be identified as implicit frames (Van der Meer, 2016). Word clusters, i.e. factors, are thus indicative of implicit frames used in the texts that formed the communication about CSR. For reasons of conciseness, the number of factors to extract was fixed on ten. The decision to extract ten factors was based on the balance between theoretical meaning and practicability. Based on the Eigenvalues, the scree plot, the proportion of explained variance and the interpretation of the word clusters, there appeared to be six factors that can be considered as coherent implicit frames. The Eigenvalues of the six factor were respectively 12.63, 9.70, 6.35, 4.90, 4.23, and 3.98, thus greater than 1. Moreover, the bend in the scree plot indicated approximately six factors. Further, the increase in the proportion of explained variance diminished after the sixth factor. Finally, the interpretation of these six word clusters was theoretical meaningful. This means that the words that load

high on the same factor could (easily) be interpreted and recognized as coherent word clusters. To maximize the amount of words included in these six factors, the number of factors to extract was subsequently fixed on six. These six factors explain together 11,6 percent of the variance in word use. In order for a word to be considered as part of a factor, the factor loading should be at least .20 (see Table B1 to Table B6 in Appendix B for the words included in the implicit frames, the accompanying factor loadings and Cronbach’s Alpha). Based on common practice in this area, a frame can be regarded as present in a text if at least three words of the particular word cluster occur in the text (see McLaren et al., 2017). Therefore, a dummy variable is created for each word within a frame, indicating either the absence or presence of a word in the text. For each frame, these values of the dummy variables are computed. This resulted in six variables that indicated the number of words of the particular word cluster present in a text. Consequently, these variables are transformed into dummy variables, with the values ‘0’ (score < 3) and ‘1’ (score ≥ 3) indicating

respectively the absence and presence of the frame in a text. The six implicit frames will be outlined below. The labeling and description of these implicit frames is based on the researcher’s interpretation of the keywords.

The first implicit frame that emerged is related to the Economic Performance & Growth. This factor consisted of keywords such as operating result, business operations, consumer, economic, consequence, growth, investments, revenue, operational, products, realization, result, strategy, improved, and profit. This implicit frame is almost identical to the economic responsibility dimension discussed above. The second implicit frame to appear was Society’s Welfare. It is related to improving the welfare of employees, communities and civil society in general, including keywords such as earth, others, labor conditions, job, better, crisis, do, economy, each other, money, give, live, people, nature, society, future, change, responsibility, world, work, and care. This implicit frame corresponds to the ethical

responsibility dimension mentioned earlier. The third implicit frame emerged is related to Environmental Protection. Co-occurring words within this frame were natural gas, biomass, fuels, greenhouse gases, CO2, sustainable, electricity, energy, energy sources, energy supply, fossil, gas, climate change, development, cleaner, emission, and worldwide. Another implicit frame that came about is related to Global Business Activities & Responsibilities. Keywords within this word cluster were business community, businesses, policy, commerce, assistance, international, social, multinationals, undertake, enterprises, developing countries,

development cooperation, organizations, and responsible. The fifth implicit frame that emerged is related to Fair & Sustainable Products. This factor included keywords such as business, consumer, sustainability, fair, factories, made, produced, customers, make,

environment, produce, products, and production. The final implicit frame that appeared concerns Educational & Charitable Programs, including keywords such as involvement, foundation, help, children, together, school, safe, and volunteers. This implicit frame seems to be in line with the philanthropic responsibility dimension discussed earlier. The presence of the identified frames was aggregated to monthly levels for both the press releases and newspaper articles.

Data collection part II: Google Trends

Based on the outcomes of the automated content analysis, data with regard to the public’s attention for (the implicit frames of) CSR was obtained through the public web facility Google Trends. Google Trends provides insights in the popularity of particular issues by indicating how often a specific search term in a certain period of time is entered relative to the total search volume. Hence, Google Trends can be used to collect time series data of the dynamics of public interest in CSR. Therefore, data drawn from Google Trends was used to determine the public’s attention for the CSR issue as well as for the implicit frames of CSR.

In order to determine the search volume with respect to CSR, appropriate search terms needed to be formulated. Compared to the collection of the press releases and newspaper articles, the opportunities with regard to data collection are somewhat limited in Google Trends. Therefore, a simplified version of the search string used in LexisNexis was applied to obtain the public’s attention for the CSR issue in the period of January 2011 to December 2015 in the Netherlands3_{. With regard to the public’s attention for the implicit frames of CSR}

in the same period and region, simplified versions of the word clusters that emerged from the automated content analysis were used4_{. Words with the highest factor loadings can be}

considered as most distinctive for the implicit frames and are, therefore, included in the search term. The number of words allowed for each search term was limited by Google and differed per frame, ranging between five and eleven words. The public’s attention for the CSR issue as well as for the identified frames were aggregated to monthly levels.

Analysis

The theoretical framework was tested with the aid of time series analysis. Especially, this study draws on the VAR (Vector Auto Regression) time-series technique. A VAR model treats all variables in the model as endogenous (Freeman, Williams, & Lin, 1989;

Vliegenthart, 2014). This implies that all variables in a VAR model possibly get affected by each other. Since there are no clear-cut theoretical expectations about the unidirectional causal relationships between the domains, VAR analyses were particularly suitable (Vliegenthart, 2014). Thus, with the aid of VAR analyses, it was investigated whether the news media, organizational PR, and the public affect each other’s CSR communication in terms of salience and use of implicit frames. The VAR analyses assessed in separate equations the current value of each of the endogenous variables based on their own past and the past of other variables (Vliegenthart & Montes, 2014).

The VAR analyses began by testing the assumption of stationarity for each time series. The assumption of stationarity implies that the mean should not depend upon the time of observation (Hollanders & Vliegenthart, 2008). To test the assumption of stationarity for each series, augmented Dickey-Fuller tests have been conducted. In all instances, the augmented Dickey-Fuller test yielded significant outcomes, which indicated that the null-hypothesis of non-stationarity can be rejected (Table C1 in Appendix C). However, in one instance, the null hypothesis of non-stationarity could be rejected only on the basis of a p-value of .095.

Therefore, a lagged dependent variable model was conducted in order to test the assumption of stationarity for this specific series. The lagged dependent variable model showed that the unstandardized coefficient of the lagged variable is significantly smaller than one (Table C2 in Appendix C). Hence, all series were stationary and did not need to be differenced.

The next step involved determining the appropriate number of lags for each VAR model by comparing a series of models with different lag lengths based on the fit statistics, including likelihood ratio tests, the Final prediction error (FPE), Akaike’s information criterion (AIC), Hannan and Quinn information criterion (HQIC), and Schwarz’s Bayesian information criterion (SBIC). Since the study consisted of 60 time points, the number of lags needed to be limited in order to prevent using too many degrees of freedom. Using too many degrees of freedom can result in inefficient estimates (Brandt & Williams, 2007). Therefore, a maximum number of three lags was allowed for the selection-order criteria. This implies that it was expected that the mutual effects of the domains occur within three months. The test statistics of the likelihood ratio tests and information criteria are shown in Table D1 to Table D7 in Appendix D. The most appropriate VAR model suggested by each test is indicated by the asterisks in the tables. With regard to the sixth frame, the fit statistics indicated none of the suggested models as appropriate. Therefore, the VAR model of the frame presence of the sixth frame was disregarded. For all other instances, the most appropriate model selected for

each VAR model was indicated by the bold text in Table D1 to Table D7 in Appendix D. Based on former steps of the analyses, various VAR models were estimated. First, a VAR model was estimated for the series of salience in each domain. Additionally, five VAR models were estimated for the series of each frame in each domain. After estimation of the VAR models, Granger-causality tests, Ljung-Box Q tests, Engle’s ARCH tests, cumulative impulse response functions (CIRF), and forecast error variance decompositions (FEVD) were obtained. Granger-causality tests were used to determine whether the domains were useful in forecasting each other. A series is Granger causing another series when the prediction of the series based on its own past improves when adding the past of the other variable in the equation (Vliegenthart & Montes, 2014). To address the issue of autocorrelation, the

(squared) residuals of the series in the VAR models were tested with the aid of Ljung-Box Q tests and Engle’s ARCH tests. Furthermore, the CIRF and FEVD were explored in order to establish the direction and size of the significant effects emerged from the Granger-causality test. The CIRF provides information about the consequences of a shock in one series at a certain time point on the subsequent values of the other series (Vliegenthart & Montes, 2014). The CIRF is, therefore, useful in gaining insight mainly in the direction (positive or negative) of the effects. On the other hand, the FEVD assesses over time the fraction of the variation in each of the endogenous series that can be attributed to the past of the own series and to the past of each of the other endogenous series (Vliegenthart & Montes, 2014). The FEVD, thus, provides information about the size of the effects.

With the aid of the above-mentioned analyses, the hypotheses were tested. Since the time points of this study were somewhat limited and some VAR models had to deal with loss of degrees of freedom, the outcomes of the study were viewed from a more lenient

Results Salience

Before estimating a VAR model for the series of salience of each domain, the series were inspected in order to get some preliminary insights in cross-time fluctuations of these series. Figure 2 shows the over-time dynamics of the salience of CSR within the news media, organizational PR, and among the public in the period of January 2011 to December 2015. In general, the issue of CSR was more salient in press releases compared to newspaper articles. The salience of CSR in newspaper articles was rather stable over the years, ranging between approximately five and 15 newspaper articles per month. The salience of CSR in press releases had more fluctuations. Although the number of published press releases regarding CSR ranged approximately between 15 and 40 press releases per month in the first few years, the salience of CSR in press releases declined slightly to a range from five to 25 press releases

Figure 2. Salience of CSR within the three domains, 2011-2015

Note. The lines for the newspaper articles and press releases were based on the number of published articles per month; the line for Google Trends Index was based on a standardized search volume index between 0 and 100.

from 2014 onwards. Hence, the difference between the salience of CSR within press releases and newspaper articles became smaller over the years. With regard to the salience of CSR among the public, there were quite some peaks and especially declines over the years. The public attention slightly decreased over the years, roughly ranging between 50 and 85 in the first few years and between 40 and 75 in the later years. In addition to the graph, the strength and direction of the correlations between the series at the same time point were inspected to check whether the series fluctuate simultaneously. The correlation between the salience of CSR in the news media and among the public was r = .09, indicating a very weak positive relationship. The correlation between the salience of CSR in the news media and

organizational PR was r = -.13, which indicated a very weak negative relationship. Moreover, the correlation between the salience of CSR in organizational PR and among the public was r = .59, which implies a fairly high positive correlation. However, this does not provide any information about the existence and strength of causal relationships (Vliegenthart & Montes, 2014). Therefore, a VAR model with lagged values of the independent series was estimated. A VAR model with two lags was estimated for the series of salience of each domain. This VAR model explained the value of each of the endogenous series of salience based on its own past and on the past of the other series. The outcomes of the estimated VAR model of salience with two lags and the Granger-causality tests are shown in Table 1. The results showed that the VAR model was successful in predicting substantial amounts of variation in the change of salience of CSR in the news media, organizational PR, and the public.

Explained variances ranged between 28.6 percent and 31.3 percent. Granger-causality tests indicated one nearly significant causal effect at the .05 level of significance and three nearly significant causal effects at the .10 level of significance. Since tendencies were explored in this study, these four effects were treated as if they were significant. Firstly, in line with hypothesis 2a, the news media did Granger cause organizational PR. Further, in accordance

Table 1. VAR Analyses and Granger-Causality Tests for the Series of Salience

The News Media Organizational PR The Public

Independent variable The News Media

L1 .361(2.85)** .003(.05) .006(.13) L2 -.096(-.76) -.052(-.89) -.088(-1.92) Constant 12.936(3.95)*** Granger-causality 5.896(.052) 4.250(.119) Organizational PR L1 .085(.27) .179(1.27) .079(.66) L2 -.732(-2.31)* .322(2.21)* -.138(-1.20) Constant 20.046(2.45)* Granger-causality .800(.670) 3.846(.146) The Public L1 -.100(-.25) .340(1.91) .290(1.91) L2 -.706(-1.76) .068(.37) -.217(-1.49) Constant 62.786(6.07)*** Granger-causality 3.772(.152) 1.593(.451) R2 _{.286 .292 .313}

Log likelihood (full model) -538.258 -538.258 -538.258 Ljung-Box Q (df=20) 17.510(.620) 14.376(.811) 19.066(.518) Engle’s ARCH (df=20) 28.423(.100) 19.838(.468) 9.453(.977)

N 58 58 58

Note: : * p < .05, ** p < .01, *** p < .001; VAR coefficients are reported with z-scores in parentheses; Granger causality chi-square statistics are reported with p-values in parentheses; Ljung-Box Q and Engle’s ARCH test statistics are reported with p-values in parentheses.

with hypothesis 1a, the news media did Granger cause the public. With regard to research question 1a, the public did Granger cause the news media. Finally, concerning research question 2a, organizational PR did Granger cause the public. The effect of organizational PR on the news media (hypothesis 2b) was not (nearly) significant, hence the hypothesis was rejected and not further explored. Moreover, the effect of the public on organizational PR (research question 2a) was not (nearly) significant, which means that the salience of CSR among the public did not affect the salience of CSR in organizational PR and this direction of the research question was not further explored. The Granger-causality tests indicated merely whether a series exerted a significant influence on another series (Vliegenthart & Montes,

2014). In order to acquire information about the size and direction of the significant effects, the CIRFs and FEVD were considered. However, before exploring the size and direction of the effects, the issues of autocorrelation and contemporaneous correlation are outlined.

Ljung-Box Q tests were conducted in order to test the null hypothesis that residuals were white noise. The tests indicated that the VAR model yielded white noise in the residuals of the news media, organizational PR, and the public (Table 1). Furthermore, Engle’s ARCH tests were estimated in order to test for autocorrelation in the squared residuals. The tests indicated that the null hypothesis of no autoregressive conditional heteroscedasticity could be rejected, hence the VAR model yielded no autocorrelation in the squared residuals of the news media, organizational PR, and the public (Table 1). Additionally, there was substantial contemporaneous correlation between the residuals of organizational PR and the public (Table E1 in Appendix E). This high contemporaneous correlation, i.e. correlation across the series within the same month, could be an indication that the salience of CSR in the domains organizational PR and the public was affected by external factors at the same time

(Vliegenthart, 2014). Moreover, it could be an indication that a lower aggregation level was desirable (Vliegenthart, 2014). Agenda-setting effects might happen more quickly, and this may be ignored in the monthly aggregated series that were used in this study. Therefore, it might be desirable to choose for weekly aggregated time series.

To establish the direction and size of the significant effects, the CIRFs and FEVD were explored. Due to the limited time points of this study, a 90 percent confidence interval has been used for both tests. Figure 3 displays the CIRFs, indicating the cumulative change in the dependent series as a consequence of a one-unit increase in the independent series. With regard to hypothesis 2a, the CIRF showed that after a five-month time span5_{, an additional}

newspaper article about CSR in the news media has resulted in a 1.37 decrease of published press releases regarding CSR in these five months (see Figure 3a). The CIRF indicated that

Figu re 3. C umul ati ve im pulse r esponse f un cti ons for the sa li enc e of CS R . No te. (a ) effe ct of ne ws m ed ia o n org an iza tio na l P R, ( b) effe ct of ne ws m ed ia o n th e pu bli c, (c ) effe ct of th e pu bl ic o n th e ne ws m ed ia, a nd ( d) effe ct of org an iza ti on al P R on th e pu bli c. CI = co nfid en ce in terv al; irf = imp ulse re sp on se fu nc ti on .

the effect of CSR salience in the news media on CSR salience in organizational PR was initially zero, but was negative from the second month onwards. The negative effect was significant after the fourth month and nearly significant at the other points in time. However, as the number of observations in this study was limited and the vast majority of the

confidence intervals was below zero at any point in time, it can quite certainly be stated that the effect of the CSR salience in the news media on the CSR salience in organizational PR was negative. Regarding the FEVD, it can be seen that after a five-month time span, 10.11 percent of the variance in the salience of CSR in press releases can be attributed to the

salience of CSR in newspaper articles (CI = [-1.60 percent, 21.83 percent]). This outcome was in line with hypothesis 2a that predicted a causal effect of the news media on organizational PR. However, contrary to the expectation, the direction of the effect was negative instead of positive. Therefore, hypothesis 2a has been rejected.

An inspection of the CIRF presented in Figure 3b suggested that after a time-span of five months, an additional newspaper article about CSR in the news media led to a 1.80 decrease of the salience of CSR among the public in these five months. This negative effect of the salience of CSR in the news media on the salience of CSR among the public was

significant from the third month onwards. The FEVD suggested that after five months shocks in the salience of CSR in the news media account for 12.86 percent of the variation in the salience of CSR among the public (CI = [-.71 percent, 26.42 percent]). The expectation formulated in hypothesis 1a was that the salience of CSR in the news media positively affect the salience of CSR among the public. Although the outcomes were in line with the predicted causal effect of the news media on the public, a negative effect was found instead of a

positive one. Hence, hypothesis 1a was rejected.

Figure 3c suggested that an one-unit increase in salience of CSR among the public after month five resulted in a .12 decrease of the salience of CSR in the news media in these

five months. The CIRF showed that the effect of CSR salience among the public on CSR salience in the news media was negative, but did not became significant at any point in time. However, the majority of the confidence intervals was below zero, which implies that it is almost certainly a negative effect. The FEVD indicated that after a five-month time span, 5.83 percent of the variation in the salience of CSR in the news media can be attributed to the salience of CSR among the public (CI = [-3.66 percent, 15.31 percent]). Hence, regarding research question 1a, the outcomes indicated that the salience of CSR among the public negatively affects the salience of CSR in the news media.

Regarding research question 2a, the CIRF presented in Figure 3d suggested that after a five-month time span, an one-unit increase in the salience of CSR in organizational PR led to a .88 increase of the salience of CSR among the public in these five months. This effect of CSR salience in organizational PR on CSR salience among the public was positive and significant at any point in time. The FEVD suggested that after a time span of five months, 29.78 percent of the variation in the salience of CSR among the public could be attributed to the salience of CSR in organizational PR (CI = [12.58 percent, 46.97 percent]). Hence, with regard to research question 2a, the results indicated that the salience of CSR in organizational PR positively affect the salience of CSR among the public. Figure 4 presents an overview of the (in)significant results of this VAR model.

Implicit frames

Initially, the over-time dynamics of the series of frame presence were inspected within the three domains in the period of January 2011 to December 2015. Figure 5 presents, for example, the series of frame presence of ‘Economic Performance & Growth’. The over-time dynamics of the other series of frame presence are presented in Figure F1 to Figure F4 in Appendix F. Furthermore, the correlations between the three series of frame presence were

Figure 4. Outcomes of the VAR model for the series of salience

Figure 5. Frame presence of ‘Economic Performance & Growth’ within the three domains, 2011-2015

Note. The lines for the newspaper articles and press releases were based on the percentage of monthly frame presence (for the clarity of the graph, newspaper articles were multiplied by 10 and press releases were multiplied by 100); the line for Google Trends Index was based on a standardized search volume index between 0 and 100.

inspected for each frame. In all instances the correlation coefficients indicated (very) weak relationships, both positive and negative (see Table G1 to Table G5 in Appendix G). This means that there were, if any, very weak associations between the series of frame presence in the three domains at the same time point.

For each frame, based on former analyses, a VAR model was estimated for the series of frame presence in the news media, organizational PR, and the public. Hence, five VAR models with the appropriate number of lags were estimated. Given that all steps of the VAR analysis were executed for each of the five frames, the results are presented in a brief manner. In this section, only the (nearly) significant effects indicated by the Granger-causality tests were discussed. These (nearly) significant outcomes of the Granger-causality tests are shown in Table 2. An overview of all outcomes is presented in Table I1 to Table I5 in Appendix I.

Granger-causality tests indicated three significant causal effects at the .01 level of significance, two at the .05 level, and two at the .10 level. Furthermore, there were two nearly significant causal effects at the .10 level of significance. Since tendencies were explored in this study, these two effects were discussed as well. Firstly, outcomes of the Granger-causality tests indicated that the news media did Granger cause the public. This was in line with hypothesis 1b and applied to the ‘Economic Performance & Growth’ and ‘Society’s Welfare’ frame. Moreover, results indicated that organizational PR did Granger cause the news media. This was in accordance with hypothesis 2d and proved to be true for the

‘Economic Performance & Growth’, ‘Society’s Welfare’, and ‘Fair & Sustainable Products’ frame. With regard to research question 1b, Granger-causality tests suggested that the public did Granger cause the news media. This was merely true for the ‘Environmental Protection’ frame. Finally, concerning research question 2b, the Granger-causality tests indicated that the public did Granger cause organizational PR. This proved to be true for the ‘Environmental Protection’, ‘Global Business Activities & Responsibilities’, and ‘Fair & Sustainable

Table 2. Overview of (Nearly) Significant Effects of VAR models of Frame Presence

The News Media Organizational PR The Public

Independent variable The News Media

Frame 1 (2 lags) Granger-causality 3.901(.142) CIRF (5) 6.163 (CI = [.175, 12.152]) FEVD (5) 6.433 (CI = [-4.900, 17.765]) Frame 2 (1 lag) Granger-causality 4.204(.040) CIRF (5) 18.936(CI = [1.690, 36.183]) FEVD (5) 6.931 (CI = [-5.421, 19.284]) Organizational PR Frame 1 (2 lags) Granger-causality 11.733(.003) CIRF (5) .606 (CI = [.225, .987]) FEVD (5) 15.128 (CI = [.649, 29.607]) Frame 2 (1 lag) Granger-causality 2.917(.088) CIRF (5) .509 (CI = [-.028, 1.047]) FEVD (5) 4.138 (CI = [-3.949, 12.225]) Frame 5 (3 lags) Granger-causality 5.876(.118) CIRF (5) -.610 (CI = [-1.378, .158]) FEVD (5) 7.601 (CI = [-4.032, 19.233]) The Public Frame 3 (2 lags) Granger-causality 9.890(.007) 5.105(.078) CIRF (5) -.018 (CI = [-.031, -.005]) -.006 (CI = [-.014, .002])

FEVD (5) 12.511 (CI = [-.045, 25.066]) 7.833 (CI = [-.2.809, 18.475])

Frame 4 (1 lag) Granger-causality 8.790(.003) CIRF (5) .011 (CI = [.004, .018]) FEVD (5) 13.586 (CI = [-.473, 27.646]) Frame 5 (3 lags) Granger-causality 8.298(.040) CIRF (5) -.003 (CI = [-.012, .005]) FEVD (5) 10.772 (CI = [-1.188, 22,731])

Note: : * p < .05, ** p < .01, *** p < .001; For Granger causality, chi-square statistics are reported with p-values in parentheses; CI = confidence interval.

Products’ frame. However, the opposite effect of organizational PR on the public (research question 2b) was not (nearly) significant for any frame. This means that the use of implicit frames of CSR in organizational PR did not affect the attention for these implicit frames among the public. In addition, the effect of the news media on organizational PR (hypothesis 2c) was not (nearly) significant for any frame. Hence, the hypothesis was rejected and was not further explored.

With regard to the autocorrelation in the (squared) residuals, the Ljung-Box Q tests and Engle’s ARCH tests indicated that there was some autocorrelation left in the residuals and squared residuals of the series of the news media and especially the series of the public (see Table I1 to Table I5 in Appendix I). This means that there was some information from the series’ own past that was not used to predict the current value (Vliegenthart & Montes, 2014). In general, the issue of autocorrelation can be solved by including additional lags in the VAR model. However, the trial-and-error process showed that in almost all instances (except one) an additional lag did not solve the problem of autocorrelation. In addition, adding extra lags to the models would result in less efficient models. Therefore, the VAR models were not

adjusted for reasons of parsimony and it was accepted that these models were not (entirely) perfect. Further, the contemporaneous correlations between the residuals of the series of frame presence were negligible, indicating that a monthly aggregation level is appropriate and desirable for framing effects (Table H1 to Table H5 in Appendix H).

Figure J1 to Figure J4 in Appendix J display the CIRFs of the significant effects. Reading examples were added to the CIRFs in order to assist the readers. Additionally, for every significant effect, the CIRFs and FEVDs after five-month time spans are presented in Table 2. Given that none of the hypotheses or research questions are fully supported by all frames, the hypotheses can only be partially confirmed and the research questions can only be answered with moderate certainty. Although not explicitly mentioned here, the hypotheses

were rejected and the research questions were answered with no effects with regard to the frames that did not have causal effects.

With regard to hypothesis 1b, the outcomes concerning the first and second frame were in line with the predicted positive effect of the news media on the public. Hence, hypothesis 1b has been partially confirmed on the basis of the first and second frame.

Moreover, the outcomes with regard to the first and second frame were in line with hypothesis 2d, which stated that the implicit frames used by organizational PR regarding CSR positively affect the use of these implicit frames by the news media. Hence, hypothesis 2d was partially confirmed on the basis of the first two frames. However, contrary to the expectation, the direction of this effect was negative for the fifth frame. Therefore, hypothesis 2d was partially rejected on the basis of the fifth frame. In addition, concerning research question 1b, the outcomes with regard to the third frame indicated that the attention for certain implicit frames of CSR among the public negatively affect the use of these implicit frames by the news media. Finally, with regard to research question 2b, the effect of the frame presence among the public on the frame presence in organizational PR was negative for the third frame, while it was positive for the fourth frame. With regard to the fifth frame, the effect was positive as well as negative. Hence, the attention for implicit frames of CSR among the public both positively and negatively affects the use of these implicit frames in organizational PR.

Discussion

The aim of this study was to examine the multidirectional causal relationships between the news media, organizational PR, and the public in terms of salience and implicit frames of CSR.

With regard to the multidirectional relationship between the news media and the public in terms of CSR salience, the findings of this study demonstrated a negative effect for both directions of this relationship. Firstly, contrary to first level agenda-setting, this implies that the more salient CSR was in the news media, the less salient CSR was among the public. Hence, hypothesis 1a has been rejected. This finding can be explained due to the high number of issues competing for public attention on a daily basis. While a lot of issues compete for public attention, only a couple of issues succeed in gaining substantial public attention

(McCombs, 2004). Although the news media play a major role in the transfer of issue salience toward the public (e.g. McCombs, 2005; Kiousis et al., 2006), it is impossible for all issues to gain extensive attention. Since the monthly salience of CSR in the news media is not

particularly high, it seems plausible that the issue of CSR was not successful in gaining public attention. In addition, this finding can be explained due to differences between Google Trends measurements and traditional measurements of the public agenda (e.g. most important

problem (MIP) questions). While answering a survey question usually has the aim of

expressing an opinion, the aim of a Google search is to find information (Mellon, 2013). With regard to CSR, there might be various reasons why people want to find information and search for it on Google. Students might search for this topic for an exam, while other people search for it based on work-related goals or personal experience. Hence, besides media attention, there could be other reasons that influence the number of Google searches for CSR. Vice versa, with regard to research question 1a, the outcomes indicated that the more salient CSR was among the public, the less salient CSR was in the news media. This implies that the news media remain silent about issues that are more prominent among the public. This study is one of the first to examine the effect of reversed agenda-setting with respect to salience of CSR. Hence, further research is needed to explore this effect.

was investigated. As opposed to agenda-building theory and previous research about news dynamics (Downs, 1972; Hollanders & Vliegenthart, 2008; Vasterman, 2005), the outcomes of this study indicated a negative effect of the news media on organizational PR in terms of CSR salience. Contrary to the expectation in hypothesis 2a, this implies that the more salient CSR was in the news media, the less salient CSR was in organizational PR. This suggests that organizations decide to remain silent about their CSR activities in times of prominent media attention for CSR. Hence, hypothesis 2a has been rejected. This outcome can, however, be well explained. Previous research indicated that media attention could be considered as a significant driver of CSR activities by organizations (e.g. Zyglidopoulos, Georgiadis, Carroll, C., & Siegel, 2012). Moreover, previous studies showed that organizations that disclose their CSR activities attract more attention and criticism from stakeholders (e.g. Ashforth & Gibbs 1990; Du et al., 2010). Hence, it can be argued that in times of prominent media publicity for CSR, organizations decide to remain silent about their CSR activities in order to avoid critical stakeholder attention. This does not necessarily mean that CSR is less prominent among organizations, instead it can be argued that organizations are more actively involved in CSR as a consequence of media attention. Despite the decrease of communication about CSR in press releases, the number of CSR-related activities of organizations might increase due to the media attention, i.e. there could be a difference between communication and action. In this regard, although not applicable to CSR communication, there can still be a first level agenda-setting effect of the salience of CSR in the news media on the salience of CSR in

organizational PR. Vice versa, as opposed to agenda-building theory and previous research about information subsidies (Davis, 2000; Deephouse, 2000; Kiousis et al., 2006), there was no causal effect of the salience of CSR in organizational PR on the salience of CSR in the news media. Hence, hypothesis 2b was rejected. The absence of this agenda-building effect can be explained due to the complex relationship between organizational PR and the news