The pains and gains of showing off CSR activities on social media

(1)

THE PAINS AND GAINS OF SHOWING OFF

CSR ACTIVITIES ON SOCIAL MEDIA

Dave Timmer – 11146915 June 22, 2017

Master Thesis Supervisor: dr. A.S. Nayak

(2)

Abstract

This study examines in what ways CSR-related messages affect both the volume and valence of customer engagement on social media. A cross-sectional quantitative data sample has been scraped from Facebook after determining the brand personality scores via a survey. The findings of the study are that CSR communication negatively affects volume and brand image, but positively affects valence of customer engagement. Additionally, volume, valence and brand image go up when the CSR activity shows a good fit with the organisation’s objectives. The study also demonstrates that while it is important for all firms to obtain a decent CSR fit, this is even more the case for companies operating in controversial industries, since obtaining a fit has a larger effect on the brand personalities of these firms.

Statement of originality

This document is written by student Dave Timmer who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Introduction

Personal and corporate goals are shifting from providing for oneself to looking after each other. In 2015, 66% of 30,000 respondents from over 60 countries said they were willing to pay more for sustainable brands, a figure that skyrocketed compared to 55% in 2014 and 50% in 2013 (Nielsen, 2015). The biggest increase in consumers’ willingness to pay (13% compared to 2014) was among companies known for being environmentally friendly or committed to social values (Nielsen, 2015). This signifies that subjects such as sustainability, public health and welfare distribution are receiving more attention globally. Organizations are expected to join this trend by setting up and communicating their initiatives with regard to corporate social responsibility (CSR) to boost public acceptance (Austin & Gaither, 2016; Du, Bhattacharya, & Sen, 2010; Engle, 2007; Welford & Frost, 2006). Examples of CSR initiatives are food producers that start to use fair trade ingredients in their products, organizations that set up poverty alleviation programs in developing countries, or companies that are actively trying to reduce their carbon footprint. This emerging concept has not been left unexamined in the research field either, where CSR has been studied frequently over the last years (Crane & Glozer, 2016).

Research suggests that when a company increases exposure of their ethical and social objectives, they are more likely to attract critical stakeholder attention (Morsing & Schultz, 2006). This increased CSR awareness has a positive effect on the firm beyond stakeholders’ consumption, by also influencing the employment and investment domains (Sen, Bhattacharya, & Korschun, 2006). However, researchers also discovered negative effects when companies proliferate unsubstantiated social and ethical claims, labelled as ‘green washing’, which result in the increase of cynicism and mistrust among its stakeholders (Jahdi & Acikdilli, 2009). The potential negative effects of CSR communication on a firms’ reputation even outweigh the

(7)

positive effects (Eberle, Berens & Li, 2013), which indicate that a firm needs to carefully evaluate their communication channels.

Although it is suggested that corporate reputation can be improved by using interactive channels to communicate about CSR (Eberle et al., 2013), researchers have paid little attention to the impact and effects of social media in the CSR literature (Kent & Taylor, 2016; Whelan, Moon, & Grant, 2013). Moreover, recent research state that the social media endeavours of a firm are a major factor in determining its CSR performance, since its stakeholders turn to social media platforms to evaluate the firms’ reputation and functioning (Bonner & Friedman, 2012). Additional studies pointed out that engaged consumers exhibit enhanced consumer loyalty, satisfaction, connection, trust, empowerment, commitment, and emotional bonding (Brodie, Ilic, Juric, & Hollebeek, 2013). This signifies that customer engagement can be an important factor for organisations in evaluating its CSR performance.

However, little is known about in what ways a firm’s communication about CSR influences customer engagement on social media (Kent & Taylor, 2016; Whelan et al., 2013); while it could prove to be interesting for both firms and academics to examine this phenomenon. For companies, understanding the factors that influence the effectiveness of their CSR communication on social media may help them to establish successful CSR campaigns; in which the heightened customer engagement due to the social media campaign may help firms to realise a better reputation (Eberle et al., 2013) and relationships with their customers (Brodie et al., 2013). For academics, studying the underexplored area of CSR communication on social media (Kent & Taylor, 2016) may help research progress towards a more complete understanding of both CSR-related communication and customer engagement on social media, which seem to be emerging fields in the literature. This is demonstrated by the extensive growth of research covering these two issues over the last years (Kent & Taylor, 2016; Tsimonis & Dimitriadis, 2015).

(8)

Considering the potential benefits of customer engagement and its presence in the online environment, it should be worthwhile measuring the relationship between a firm’s CSR communication and customer engagement in the social media environment. Given that there might be multiple factors that jointly determine the effect on customer engagement, this study examines the following research question:

RQ: In what ways does an organisation’s communication about CSR initiatives influence customer engagement on social media?

The researcher is aware of the broadness of the research question, but this is deliberately done to be able to dive into different domains regarding CSR communication and customer engagement. In the literature review, sub-questions have been formulated to examine each domain specifically. These sub-questions test the effects of a firm’s CSR fit, brand image and industry type regarding CSR communication and customer engagement. Consequently, the research design elaborates on the two-staged data collection and analysis, where a survey is used for the first and a web scraper for the second stage. After elaborating on the results of this analysis, these results are linked to the literature and the final conclusions are revealed.

(9)

Literature review

The review of the relevant literature for this study starts with a broad perspective, and zooms in on each specific element necessary to answer the research question. First, there needs to be consensus on the definition of CSR used in this study. Thereafter, the recognized effects of CSR communication are evaluated, which is also done for customer engagement to be able to formulate the first hypothesis. Subsequently, literature on CSR fit, brand personality and industry type relating to customer engagement is reviewed, and hypotheses are formulated for each of the three sub-sections.

Defining corporate social responsibility

To be able to study these diverse effects that CSR communication may have on its stakeholders, it is important to reach a consensus on the definition of the focal concept of CSR. Dahlsrud (2008) analysed 37 definitions of CSR to constitute an unbiased and clear definition. He concludes that there are five dimensions within CSR definitions that are consistently referred to: the stakeholder, social, economic, voluntariness and environmental dimensions. A proper CSR definition does not necessarily have to incorporate all five of these dimensions, but in 97% of the definitions Dahlsrud (2008) examined, at least three of these dimensions are included. The definition from the Commission of European Communities, which incorporates all five dimensions, has the highest frequency count on Google (Dahlsrud, 2008). They define CSR as: “A concept whereby companies integrate social and environmental concerns in their business operations and in their interactions with stakeholders on a voluntary basis” (European Commission, 2001, p. 8).

Other researchers suggest that an important factor in defining CSR is the voluntary commitment a firm undertakes to act socially responsible beyond its direct economic technical interest, thereby exceeding the expectations of common corporate behaviour imposed by

(10)

society (Carroll, 1991; Hou & Reber, 2011; Mohr, Webb & Harris, 2001). Altogether, CSR is defined as: ‘a firm’s voluntary commitment in the social, environmental, economic and/or stakeholder domain to act socially responsible beyond its own direct economic or technical interest.’ The definition is used throughout this study to assure consensus on the CSR concept.

CSR communication

In 1863, economist Adam Smith introduced the concept of the invisible hand of the marketplace, in which he argued that when business owners want to produce the greatest social good, they should merely focus on their own profits. Whether corporate decision makers should be concerned with anything other than generating profits has since remained a major question in business (Mohr, Webb & Harris, 2001). In recent studies, researchers claim that acting socially responsible can also maximize a firm’s business returns (Du et al., 2010). This shifts the focus of even the most profit-oriented organisations towards initiatives of corporate social responsibility.

Of course there are different perspectives on corporate social responsibility, ranging from the profit-oriented focus in which CSR is another instrument to maximize shareholder value, to the intrinsically motivated responsible firm in a dynamic social system, and everything in between (Sen & Bhattacharya, 2001). Because of these differences in underlying motives for CSR, stakeholders may react differently to various initiatives a firm undertakes. Stakeholders will generally perceive a company’s motive for CSR as either extrinsic or intrinsic (Du et al., 2010). When the firm is perceived as attempting to increase its profits using CSR, this is an extrinsic motivation; but when it is perceived as a genuine action out of concern for the focal issue, it is perceived as an intrinsic motivation. Generally, CSR activities that stakeholders attribute to intrinsic motives lead to positive evaluations, while extrinsic motives may dampen the positive or even lead to negative stakeholder evaluations of the firm (Du et al., 2010; Yoon, Gurhan-Canli, & Schwarz, 2006).

(11)

Eberle et al. (2013) argue that using interactive channels to communicate CSR can boost a firm’s reputation and word-of-mouth, since enhanced interactivity increases message credibility and leads to stronger feelings of identification with the firm. Other researchers also found that communication of CSR policies and activities impacts a firm’s reputation and brand image (Arendt & Brettel, 2010; Calabresea, Costaa, & Rosati, 2015; Jahdi & Acikdilli, 2009; Morsing & Schultz, 2006). Communicating CSR activities may also drive customer loyalty and turn customers into brand ambassadors, or even cause customers to seek employment with the firm (Du et al., 2010). Yet, due to green washing attempts and extrinsic motives for CSR, all firms face the challenge of increasing consumer scepticism, mistrust, and negative user evaluations, which makes it more and more difficult to convince them of the firm’s honest intentions (Calabresea et al., 2015; Dawkins, 2005; Du et al., 2010; Eberle et al., 2013; Jahdi & Acikdilli, 2009; Parguel, Benoit-Moreau, & Larceneux, 2011). Nevertheless, consumer interest in ethical products and companies is increasing (Mohr et al., 2001; Nielsen, 2015). This indicates that communicating CSR policies in the right way will become increasingly important for firms to keep their customers up-to-date with their CSR initiatives, and maintain a proper reputation and image (Jahdi & Acikdilli, 2009).

It is not easy for firms to develop a clear CSR communication strategy, due to the diverse communication channels available and the different information requirements of the different stakeholders (Dawkins, 2005). This signifies that each channel needs to be examined individually to be able to address the specific challenges of communicating CSR activities to consumers. Using social media is considered to play a major role in determining a firm’s CSR effectiveness, since consumers often turn to social media to evaluate a firm’s functioning (Austin & Gaither, 2016; Bonner & Friedman, 2012); and it is an effective channel to build relationships with consumers (Austin & Gaither, 2016; Etter, 2013). It is even suggested that social media can help to accelerate and expand the global change towards a sustainable society

(12)

(Rivera-Arrubla & Zorio-Grima, 2016). However, in the extensive literature field on CSR, the use of social media as a CSR communication channel has received little attention (Austin & Gaither, 2016; Kent & Taylor, 2016; Rivera-Arrubla & Zorio-Grima, 2016; Whelan et al., 2013). This indicates that there is a need to examine the role of social media in CSR communication, and proposes an opportunity to help the field progress towards a more complete understanding of this phenomenon.

Customer engagement

In the social media environment, customer engagement is often viewed as the result of effective communication (Sashi, 2012). Engagement behaviours are the behavioural manifestation of motivational drivers towards a firm beyond transactions. These behaviours include word-of-mouth (WOM) activities, recommendations, writing reviews, blogging, helping other customers, and even engaging in legal action (Van Doorn et al., 2010). Many of these activities appear to be exhibited in online environments, of which social media is the most prominent. In the social media environment, customer engagement is defined as “taking some action beyond viewing or reading” (Delahaye Paine, 2011, p. 60). For instance, this may be when customers are liking, commenting, or sharing a post. A change in the volume of customer engagement does not necessarily have a positive effect on the firm, since customers may also engage in a negative manner. For example, by placing comments of disappointment or frustration towards a company, which may have reputational consequences for the focal firm (Van Doorn et al., 2010). This signifies that not only the volume of customer engagement, but also the valence (i.e. positive or negative) of this engagement is of importance to firms (Van Doorn et al., 2010).

Engagement is a process with distinct states, which implies that the intensity of customer engagement differs over time (Brodie et al., 2013). This indicates that companies need to do their best to keep customers engaged. The engagement process consists of a range of

(13)

sub-processes that reflect the interactive experience customers have with other customers within online communities, and the value that they create together (Brodie et al., 2013). The engagement behaviour of customers has multiple consequences for a firm. Behaviour such as WOM, referrals and generating information may influence purchase behaviour of multiple customers, thereby affecting the financial performance of a firm. Moreover, these behaviours have reputational consequences which affect brand awareness and image of a firm (Van Doorn et al., 2010). If a firm succeeds in engaging customers, they may expect an increase in the loyalty, satisfaction, connection, trust, empowerment, commitment, and emotional bonding of these customers (Brodie et al., 2013).

Since researchers have demonstrated the importance for firms to engage their customers, and the importance of the sentiment of this engagement, the effects of CSR communication on both the volume (e.g. amount of likes, comments or shares) and valence (e.g. positive or negative sentiment) of customer engagement is being studied. This leads to the first sub-question: In what way does a CSR-related post influence customer engagement on social media?

Previous research has mainly focused on the beneficial effects of CSR activities on customer engagement (Mattila, Wu & Choi, 2016), but to reap the benefits of these effects, firms must continuously communicate their CSR commitment and effort (Calabresea et al., 2015; Du et al., 2010; Morsing & Schultz, 2006). Researchers also identified multiple factors that reduce the effectiveness of CSR activities on customer engagement (Mattila et al., 2016). However, since the majority of prior research emphasizes the positive effects of CSR activities on customer engagement, a positive relation between communicating CSR activities and customer engagement is expected.

(14)

Austin & Gaither (2016) found that when comparing post topics, posts containing CSR were received the most favourably on their valence scale, which indicates a positive relation between CSR communication and customer engagement valence.

◊ H1b: CSR communication leads to an increase in the valence of customer engagement.

CSR fit

The characteristics of the specific CSR activity and its underlying motives should also be considered when evaluating effectiveness; since stakeholders may react differently to various initiatives a firm undertakes. In a comprehensive summarization of CSR initiatives, Sen & Bhattacharya (2001) identified six CSR domains: community support, diversity, employee support, environment, overseas operations (human rights), and product. Comparing these domains with other relevant literature, and consequently to this paper’s definition of CSR (‘a firm’s voluntary commitment in the social, environmental, economic and/or stakeholder domain to act socially responsible beyond its own direct economic or technical interest’); the following relevant domains of CSR activities have been identified: employee support, environment, community support, diversity, human rights (Hou & Reber, 2011; McWilliams & Siegel, 2000; Moir, 2009; Roberts, 1992; Sen & Bhattacharya, 2001).

While the firm’s motives are important factors in understanding customer evaluation of a company’s CSR initiatives (Du et al., 2010; Yoon et al., 2006), the relation of the activity with its business interests also play a role. The effectiveness of such an initiative will be reduced if the CSR activity seems to be acting against the company’s interest (Austin & Gaither, 2016). For example, when Shell starts an initiative to reduce carbon emissions, since they are being held accountable for a lot of emissions themselves. Stakeholders expect firms to pursue social initiatives that have a good fit or logical association with their business activities (Du et al., 2010). Hence, it is important for firms to demonstrate the CSR fit when starting a new initiative,

(15)

to show that this initiative is both beneficial to their business and society (Calabresea et al., 2015; Porter & Kramer, 2007). Even when there is no natural CSR fit, a firm should still elaborate on its reasons for pursuing the initiative to enhance the fit perceived by stakeholders (Du et al., 2010).

Austin & Gaither (2016) acknowledge that the role of congruence between a firm’s CSR activity and its business interests need to be examined further. The fit between the business and a CSR initiative seems to be an important factor in determining the effectiveness of CSR communication, which leads to the next sub-question: In what way does the type of CSR activity a firm undertakes influence customer engagement in communicating CSR activities on social media?

Thus, it may be expected that simply communicating about CSR initiatives is not enough to engage customers, but communicating about CSR initiatives that fit with the firm’s objectives should positively influence both the volume and valence of customer engagement.

◊ H2a: Obtaining an understandable fit between a firm’s CSR initiatives and its business interests positively influences the volume of customer engagement.

◊ H2b: Obtaining an understandable fit between a firm’s CSR initiatives and its business interests positively influences the valence of customer engagement.

Brand personality

Firms need to communicate CSR initiatives effectively to be able to build brand image (Crane & Glozer, 2016). CSR communication triggers the corporate-image-building process, but this is contingent on the firm’s size, industry and marketing budget (Arendt & Brettel, 2010). Firms should thoughtfully consider what to post on social media, since what they post determines how they are being perceived by the public. Consequently, the way the firm is

(16)

perceived by the public influences the effectiveness of CSR communication, but this relation needs to be examined in further research (Mishra & Mohanty, 2013).

Aaker (1997) developed a brand personality framework, in which firms can be perceived as either sincere, exciting, competent, sophisticated or rugged. Each dimension has a few facets, and each facet has a few traits which help to determine which brand personality a company adheres to. It would be interesting to test whether a difference exists in the way different dimensions of brand personality influence customer engagement, since this has not been examined before. Previous research indicates that CSR communication may influence a firm’s brand personality, which then has a mediation effect on the relationship between CSR communication and customer engagement. Therefore, the following sub-question is: In what way does a firm’s brand personality influence customer engagement in communicating CSR activities on social media?

To be able to test the effect of a brand personality on customer engagement, a minimal of two different dimensions are needed, and not all five dimensions need to be incorporated. When evaluating the different brand personalities, sincerity and competence were found to be most in line with the setup of this study, and have thus been chosen as the two brand personality dimensions. The rationale behind the choice for these two is that the traits connected to sincerity seem to have the best fit with social behaviour (due to traits such as honest, pure, wholesome, sentimental and friendly), whereas it seems that the traits of the competence dimension best fit profit-oriented behaviour (due to traits such as technical, corporate, successful and leader). It is expected that CSR communication influences the sincerity of a brand positively, since it shows more social behaviour. The CSR fit is also expected to positively influence sincerity, because it shows even more genuine social behaviour. Sincerity is expected to have a positive influence on both the volume and valence of customer engagement, since consumers will perceive the firm in a more positive way. This leads to the following hypotheses:

(17)

◊ H3a: CSR communication on social media increases the perception of a brand as being sincere.

◊ H3b: Obtaining an understandable fit between a firm’s CSR initiatives and its business interests increases the perception of a brand as being sincere.

◊ H3c: An increase in the perception of a brand as being sincere positively influences the volume of customer engagement.

◊ H3d: An increase in the perception of a brand as being sincere positively influences the valence of customer engagement.

On the other hand, the company shows less corporate and technical behaviour when communicating CSR initiatives, which feeds the expectation that people may perceive the firm as being less competent, and will negatively impact the valence of customer engagement. However, if the firm establishes a proper fit between CSR activities and its business objectives, this may lead consumers to perceive the firm as being more competent, since it shows the leadership characteristics of the firm. Consequently, it may positively impact the volume and valence of customer engagement. This leads to the following hypotheses:

◊ H4a: CSR communication on social media decreases the perception of a brand as being competent.

◊ H4b: Obtaining an understandable fit between a firm’s CSR initiatives and its business interests increases the perception of a brand as being competent.

◊ H4c: An increase (decrease) in the perception of a brand as being competent positively (negatively) influences the volume of customer engagement.

◊ H4d: An increase (decrease) in the perception of a brand as being competent positively (negatively) influences the valence of customer engagement.

(18)

Industry type

In their study on the effect of external factors on CSR effectiveness, Arendt & Brettel (2010) found significant effects of a firm’s industry, namely that the effects of CSR on brand image and firm performance varied significantly by industry, but also by company size. Additionally, Bhattacharya & Sen (2004) found that belonging to a certain industry negatively influences the effect of CSR initiatives. For instance, in the tobacco, oil, or alcohol sector, the public tends to make unfavourable and sceptical attributions towards the CSR activities initiated by those firms. These firms are operating in the so called controversial industries (Cai, Jo & Pan, 2012). Controversial industries may be characterized by political pressures, moral debates, or social taboos. They include sinful industries like adult entertainment, gambling, tobacco and alcohol, but also comprise industries which are involved in social, environmental or ethical issues, such as the weapon, oil, cement, biotech and nuclear industries (Cai et al., 2012). Because of the negative effects the companies within these industries already have on the ethical, social or environmental domain in the eyes of the stakeholders, suspicion of these companies enhances when they try to act socially responsible (Du et al., 2010).

Although there has been some research on the relationship between industry type and CSR, comparing controversial with non-controversial industries with regard to CSR is not yet fully understood (Cai et al., 2012; Yoon, Gurhan-Canli & Schwarz, 2006). Moreover, none of the studies examined the effect on the volume or valence of customer engagement, nor did they study the effect in the social media environment. Therefore, the following sub-question will be examined in this study: In what way does a firm’s industry influence customer engagement in communicating CSR activities on social media?

It is expected that firms who operate in controversial industries experience hinder of their industry image when communicating about CSR activities. Moreover, these firms will be perceived to be less sincere and competent. On the other hand, firms who are operating in

(19)

non-controversial industries are expected to experience a boost in both volume and valence of customer engagement due to a better industry image. As a result, these firms will be perceived to be more sincere and competent. This leads to the last couple of hypotheses:

◊ H5a: Being active in a (non-)controversial industry negatively (positively) influences the effect of CSR communication on the volume of customer engagement.

◊ H5b: Being active in a (non-)controversial industry negatively (positively) influences the effect of CSR communication on the valence of customer engagement.

◊ H5c: Being active in a (non-)controversial industry negatively (positively) influences the effect of CSR fit on the volume of customer engagement.

◊ H5d: Being active in a (non-)controversial industry negatively (positively) influences the effect of CSR fit on the valence of customer engagement.

◊ H5e: The effect of CSR communication on sincerity is weaker when a firm is active in a controversial industry.

◊ H5f: The effect of CSR communication on competence is weaker when a firm is active in a controversial industry.

◊ H5g: The effect of CSR fit on sincerity is weaker when a firm is active in a controversial industry.

◊ H5h: The effect of CSR fit on competence is weaker when a firm is active in a controversial industry.

(20)

Conceptual model

The hypotheses formulated in the literature review are illustrated in the conceptual model, exhibited in Figure 1.

(21)

Research Design

To be able to do a proper study on the effectiveness of CSR communication on social media, both quantitative (i.e. number of likes, comments, and shares) and qualitative (i.e. company posts) data has been collected from Facebook. This way, real data can be obtained and used for further analysis. However, before this data was obtained, a survey was distributed to be able to measure the brand personalities of a pre-specified amount of selected companies, to ensure that these companies differ in the way the public perceives them. The survey also provided an opportunity to measure the public’s opinion on the fit of CSR activities for each company.

The sequence of data collection entails that the social media data can only be collected after the initial survey results are obtained. Therefore, the research design section, and its consecutive results section, have been divided into two parts – research design and results from the survey, followed by research design and results from the Facebook data – to maintain a logical structure throughout this research paper. Details about the population, sampling methods, and data manipulation will be elaborated upon in the different paragraphs of this chapter.

Research design – Survey

The bottleneck in the data collection process is determining the brand personality and the CSR fit, since these are measured by a survey, for which respondents are needed. The more companies that are included in the study, the longer the survey will be, which will make it more difficult to reach a sufficient number of respondents. Therefore, the amount of companies that can be selected for the study is limited.

Researchers found that firm size has an influence on the effectiveness of CSR communication (Arendt & Brettel, 2010). However, the authors did not account for the level of

(22)

familiarity the firms have among the public. In a Korean study, researchers examined the effect of corporate fame when communicating CSR to the larger public, and found that globally or nationally well-known companies generated significantly more cynical responses than individuals or not so familiar or foreign companies (Cho & Hong, 2009). The latter group generated significantly more positive responses than the globally and nationally well-known firms. Whereas it seems that familiarity level is related to company size, and both may influence the effectiveness of CSR communication on customer engagement, this study only includes large companies with widespread familiarity. To achieve this, only organisations that are included in the global Fortune 500 will be selected. This also heightens the possibility that respondents are aware of the actions of these companies, and thus perceive them in a certain way. Furthermore, firms that do not (or barely) engage in CSR communication and activities on Facebook are also included in the analysis to make a fair comparison between firms.

Industry type

To ensure an appropriate industry classification of these companies, the International Standard Industrial Classification (ISIC) of All Economic Activities by the United Nations Statistics Division has been used (4th revision). Although there are multiple active industry classifications, this one has been selected due to its global focus and the recency of the updated version (“Detailed structure and explanatory notes: ISIC Rev.4,” 2008). The sample of companies need to include a mix of companies active in controversial and non-controversial industries; this has been taken into account during the selection process. The ISIC codes of each of the companies are taken from financials.morningstar.com (2017). The selected companies are shown in Table 1.

Two independent raters test whether this industry classification is valid. A firm can operate in either a controversial or non-controversial industry, these will be scored with either a 1 or a 0, respectively. Controversial industries include adult entertainment, gambling, tobacco,

(23)

alcohol, weapon, oil, cement, biotech and nuclear industries (Cai et al., 2012); all other industries are classified as non-controversial. Complete consensus is required between the two raters. When this is not accomplished instantly, the raters will go in discussion to determine the industry. This consensus is necessary due to the limited amount of companies in the sample. Cohen’s Kappa is a statistical indicator of agreement (between -1 and 1) when comparing two independent ratings about the same observations (Landis & Koch, 1977). For the industry classification, there was perfect agreement between the two independent raters, κ = 1.000 (95% CI, 1.000 to 1.000), p < .01. Therefore, these companies and their industries are found suitable for further analysis. Six out of the fourteen companies have found to be active in controversial industries.

Table 1: Fortune 500 (2016) companies selected for analysis

Brand personality

With regard to brand personality, this study examines whether brands perceived as either ‘sincere’ or ‘competent’ differ in the effectiveness of CSR communication on social media. To classify brands in one of either category, a survey is distributed. The traits that may

Company Industry Classification (ISIC) Fortune 500 position Controversial Royal Dutch Shell Extraction of Crude Petroleum (0610) 5 Yes

Exxon Mobil Extraction of Crude Petroleum (0610) 6 Yes

Toyota Motor Manufacture of Motor Vehicles (2910) 8 No

Apple Manufacture of Communication Equipment (2630) 9 No

BP Support Activities for Petroleum and Natural Gas Extraction (0910)

10 Yes

Alphabet Web Portals (6312) 94 No

ING Group Other Monetary Intermediation (6419) 117 No

Unilever Manufacture of Soap and Detergents, Cleaning and Polishing Preparations, Perfumes and Toiled Preparations

147 No

The Coca-Cola Company Manufacture of Soft Drinks; Production of Mineral Waters and Other Bottled Waters (1104)

206 No

Anheuser-Busch InBev Manufacture of Malt Liquors and Malt (1103) 211 Yes

Nike Manufacture of Footwear (1520) 343 No

Air-France/KLM Passenger Air Transport (5110) 363 No

Philip Morris International Manufacture of Tobacco Products (1200) 398 Yes Heineken Holding Manufacture of Malt Liquors and Malt (1103) 459 Yes

(24)

determine whether a firm belongs to either category were already established by Aaker (1997), and the same will be used for this study; see Table 2.

Table 2: Brand personality traits

CSR fit

The fit between the company’s objectives and a specific CSR activity needs to be determined. By asking the survey respondents how they think a firm is doing in one of the five CSR initiative domains (employee support, environment, diversity, community support and human rights), the degree of fit between the firm and its CSR activities can be determined. For example, if a company is pursuing a human rights initiative and the respondents perceive the firm to be doing well regarding human rights, the firm has a good fit with its objectives.

Data collection

A survey is used to collect data on the brand personality scores of companies, as well as a measure to determine their CSR fit later in the data collection process. This survey has been created and distributed via Qualtrics.com. This website provides a platform to easily compile a survey, distribute the survey via anonymous links, and export the collected data in to a file of choice.

Each company section started with a short introductory text about the company, to secure that the respondents did not answer the questions with the wrong firm in mind. For each of the companies included in the study, respondents were asked whether they found that a group of traits (see Table 2) fits the image they have of that company. They are presented with six

Factor Name Facet Name Traits

Honest Honest, sincere, real Wholesome Wholesome, original, pure Cheerful Cheerful, sentimental, friendly Reliable Reliable, hard working, secure Intelligent Intelligent, technical, corporate Successful Successful, leader, confident Sincerity

(25)

groups of traits in total, of which three relate to the brand personality sincerity, and the other three to competence. Consequently, the respondents were asked how well they thought that the firm was doing in each of the five CSR domains. The order of the companies was randomized to control for any effects that might occur due to the order of presenting them.

Results – Survey

Descriptive statistics

Of the 173 respondents, 53 only opened the link, viewed the introductory page of the survey, and left the page. These 53 respondents have logically been deleted from the data file. There were 3 respondents who had completed the survey, but where no responses have been recorded for. It seems that these respondents have clicked through to the end of the survey, and consequently closed it without filling it in; these respondents have been deleted as well. Of the remaining 117 respondents, 87 have completed the survey, and 30 have partially filled in the survey. Because they did not finish the survey, there is no demographic data on these 30 respondents, but their responses are included since their answers on the questions they filled in are valid.

Of the respondents for which their demographics are known, 51.7% are male. The youngest respondent is 21 and the oldest 69 years old, but 79.1% of the sample is below 30 years old. The main reason for this is that the social network of the researcher in mostly in this age group. 52.1% of the respondents are Dutch, various nationalities made up a total of 18% of the respondents, and of the other 29.9% the nationality is unknown. The respondents are quite highly educated in general, with 90.8% following or having finished a University or University of applied sciences degree. Furthermore, more than half (51.2%) of the respondents are students, 38.4% are fulltime of part-time employees, 8.1% are entrepreneurs and the rest is either retired of unemployed.

(26)

Reliability and validity

To confirm the findings of Aaker (1997) that the different traits indeed measure brand personality, I performed an analysis to test the construct validity, i.e. whether the items do indeed measure the construct they are intended to measure (Cronbach & Meehl, 1955). Also, I tested the reliability of the results to see to which extent the data collection methods yield consistent findings. For the three trait groups that measured sincerity, a Cronbach’s alpha of .861 indicates that there is a high consistency among items in the survey (above .70, Table 3) (Cortina, 1993). Moreover, the Cronbach’s alpha would be lower when any of the three items were deleted, which indicates that I should not remove any of the three trait groups when measuring sincerity. The corrected item-total correlations are all well above .30, which indicate that all the items have a good correlation with the total score of the scale. The other three items (that measured competence) also showed a high internal consistency with a Cronbach’s alpha of .785 (Table 4). The Cronbach’s alpha would be slightly higher (.812) when the first of these three items were to be deleted, however, this would not substantially affect the reliability. The corrected item-total correlation is sufficient (.546), which means that this item has a good correlation with the total score of the scale. These findings are consistent with the research by Aaker (1997) and show that the six items used to measure the two brand personalities are both valid and reliable.

Table 3: Reliability and item-total statistics sincerity

N of items Cronbach's alpha

Cronbach's alpha based on standardized 3 .861 .861 Traits Corrected item-total correlation Cronbach's alpha if item deleted

Honest, sincere, real _.743 _.799

Wholesome, original, pure .771 .773

(27)

Table 4: Reliability and item-total statistics competence

Subsequently, I can use the six items that measure both sincerity and competence to compute a sincerity and competence score for each company. I do this by computing the mean of the three items that measure sincerity for each firm, and repeat this process for competence. Each company now has two values, one score for their sincerity, and one for their competence. These values will be used in the following analysis to determine whether there are significant differences between them.

Since there is only one independent categorical (nominal) variable (company) and one dependent interval variable (brand personality score) per analysis, a one-way univariate ANOVA is used. There is a statistically significant effect of companies on sincerity score, F (13, 1098) = 34.805, p < .01. Tukey post-hoc tests shows that the sincerity scores of Shell, Exxon Mobil, BP and Philip Morris are significantly higher than each of the other companies (p < .01 for all comparisons), but there were no significant differences between the sincerity scores of these four companies. The same goes for the scores of the other ten companies that were significantly lower (see Table 5).

There also is a statistically significant effect of companies on competence score, F (13, 1113) = 18.556, p < .01. The Tukey post-hoc tests show that there are significant differences between companies, but the results are not as dichotomous as with the sincerity scores, where the companies could be divided into exactly two groups. However, significant differences between companies are present, and shown in Table 6.

N of items Cronbach's alpha

Cronbach's alpha based on standardized 3 .785 .793 Traits Corrected item-total correlation Cronbach's alpha if item deleted Reliable, hard working, secure .546 .812 Intelligent, technical, corporate .677 .657 Successful, leader, confident .669 .664

(28)

Table 5: Post-hoc Tukey test sincerity

Table 6: Post-hoc Tukey test competence

It should be noted that a higher sincerity score indicates that respondents thought these companies were less sincere, and the same goes for competence. This signifies that the two brand personality Likert-scale scores are counter-indicative and need to be recoded, to enhance the interpretability of the data. So, a score of one becomes a seven, and the other way around. Therefore, the following formula is used to recode the brand personality scores: Brand personality score = 7 – (score – 1).

Companies N 1 2 Heineken 88 2.7803 Google 75 2.8489 Nike 93 2.9677 Apple 91 3.0659 Toyota 81 3.1605 Air France-KLM 86 3.2481 Unilever 84 3.4048 ING 88 3.4545 Coca-Cola 90 3.4815 AB InBev 45 3.4815 Shell 84 4.8690 BP 70 5.0238 Exxon Mobil 59 5.0621 Philip Morris 78 5.2308 Sig. .069 .923

Note: means for groups in homogeneous subsets are displayed Subset for alpha = 0.05

Companies N 1 2 3 4 5 6 Google 78 1.9615 Apple 92 2.0145 2.0145 Heineken 87 2.3142 2.3142 2.3142 Nike 94 2.4894 2.4894 2.4894 2.4894 Unilever 86 2.5194 2.5194 2.5194 2.5194 Toyota 81 2.5885 2.5885 2.5885 ING 88 2.6705 2.6705 Coca-Cola 92 2.6812 2.6812 Air France-KLM 86 2.7287 2.7287 AB InBev 47 2.8723 2.8723 Shell 87 3.0153 3.0153 BP 69 3.5845 3.5845 Exxon Mobil 60 3.6000 3.6000 Philip Morris 80 3.7958 Sig. .124 .097 .123 .191 .083 .997

Note: means for groups in homogeneous subsets are displayed

(29)

Subsequently, a one-way univariate ANOVA is used to test whether these is a significant difference between the companies regarding their scores on the five different CSR domains. There are statistically significant effects of companies on employee support (F (13, 939) = 7.541, p < .01), environment (F (13, 1016) = 25.795, p < .01), community support (F (13, 979) = 17.124, p < .01), diversity (F (13, 960) = 13.624, p < .01) and human rights (F (13, 866) = 14.144, p < .01). Since these items are counter-indicative, these Likert-scale domain scores are also recoded to enhance interpretation, using the following formula: CSR domain score = 5 – (score – 1).

The findings from the survey indicate that the selected companies differ significantly regarding their brand personalities, based on both their sincerity and competence scores. Furthermore, it is demonstrated that the way the public perceives a firm to do well in a certain CSR domain differs significantly between companies. This entails that these seven scores (one for each brand personality, and one for each of the CSR domains) will be used in subsequent analysis, to be able to determine whether brand personality or CSR fit influences the volume or valence of customer engagement.

Research design – Facebook

Data collection

Now that the results of the survey provide sufficient grounds to continue with the study, data on the other variables are collected through a web scraper. I chose to collect data from Facebook, since it’s the most popular social media platform where firms engage with their customers. Data is collected on a per post level to be able to compare customer engagement across different CSR (and no CSR) posts. A selection of fourteen Fortune 500 companies from different industries has been made, and only posts from these companies are obtained. The posts

(30)

need to be made on verified company pages. When a page or profile has been verified, Facebook has confirmed that this is the authentic page or profile of a company or brand (Facebook, 2017).

Unfortunately, when examining the verified Facebook pages of each company, a problem occurred. Apple’s Facebook page does not contain any posts and is therefore excluded from further analysis. Furthermore, as Air France and KLM each have their own Facebook pages, data is collected from both pages to match the survey results, which were based on the organisation Air France-KLM. Moreover, each company either has an international or Dutch Facebook page, to which you are automatically directed, but ING has both an international and Dutch page that is accessible. Data is collected from both pages to foster comparison.

Collecting data through the Facebook API allows users to collect Facebook data in a controlled way, by obtaining an API key for a limited time frame. Real-time raw Facebook data can be obtained via this approach, which enhances the reliability of the study. The data collection as well as the consequent sentiment analysis was done using R and RStudio. This software-package provides a convenient and easy-to-operate interface that allows for a thorough and precise data collection and analysis. The API key obtained from Facebook can be added in the few lines of code needed for the data collection, which fully automates the data collection process.

In the first data dump, the latest 100 posts of each of the remaining thirteen firms are collected. Thereafter, the initial classification will take place. If not more than 20 posts per company are CSR-related, another sample of 100 posts is collected for the companies that do not have a minimum of 20 CSR-related posts. This second dump only includes posts that were posted prior to the date of the first dump, to avoid any mutations due to recent events. This process continues until each firm has at least 20 CSR-related posts, or 300 posts have been classified, or the entire history of posts has been classified.

(31)

CSR communication

The main independent variable is CSR communication, which is coded by two independent raters. If a post contains a CSR message, it is classified as CSR communication and will get the value 1, otherwise this value will be 0. To be classified as CSR communication a post needs to be reasonably in accordance with the definition of CSR: ‘a firm’s voluntary commitment in the social, environmental, economic and/or stakeholder domain to act socially responsible beyond its own direct economic or technical interest.’ This classification is assessed by two raters, where an almost perfect interrater reliability is desirable (κ > .80; Landis & Koch, 1977).

CSR fit

The type of CSR activity may be classified in one of the following domains: employee support, environment, community support, diversity, human rights, other or none. See Table 7 for the descriptions and examples of each of the CSR domains.

Table 7: Descriptions and examples of the five CSR domains

CSR domain Description Examples

Employee support

“fostering safe and respectful workplaces for employees, and improving their working experience.”a

“concern for safety, job security, profit sharing, union relations, employee involvement”b

Environment

“focused on reducing a company’s overall impact on the environment in terms of climate change, energy efficiency, waste reduction and recycling.”a

“Environment-friendly products, hazardous waste-management, use of ozone-depleting chemicals, animal testing, pollution control, recycling”b

Community support

“being a responsible and good corporate citizen to the communities you serve.”a

“Support of arts and health programs, educational and housing initiatives for the economically disadvantaged,

generous/innovative giving”b Diversity

“promote diversity”a_{and “encourage and} develop current employees from diverse backgrounds.”a

“sex-, race-, family-, sexual orientation-, and disability-based diversity record and initiatives”b

“overseas labour practices [including sweatshops], operations in countries with human rights violations”b

“human rights issues such as health, poverty and education”c

a

Hou and Reber (2011, p. 167). bSen and Bhattacharya (2001, p. 226). cEngle (2007, p. 18).

Human rights “policies and actions in protecting human rights.”a

(32)

The classification of CSR types is assessed by two raters, where an almost perfect interrater reliability needs to be obtained (κ > .80; Landis & Koch, 1977). This classification is necessary to be able to determine the CSR fit or a certain post.

To determine CSR fit, the CSR domain score of a company is used when a company post is about this domain. For example, when AB InBev makes a CSR post on the diversity domain, and the company scores 2.73 on this specific domain, the CSR fit score is 2.73. Consequently, CSR fit is classified as either having a ‘good fit’ or a ‘bad fit’ with the company. In the survey, respondents were asked how well they thought the company was doing in a certain CSR domain. The recoded score of 2 represents the answer ‘slightly well’ and the score of 3 stand for ‘moderately well’. When examining the distribution of values, the cut-off point of 2.5 has been chosen, where values below 2.5 in a certain domain indicates a bad fit with the company, whereas values of 2.5 and higher indicate a good fit between the CSR initiative and the firm. When a post does not contain CSR communication, no value is listed for CSR fit. This is done to avoid positively skewing this variable, which would be done when the value of zero would be appointed to CSR fit in this case.

Customer engagement – Volume

The dependent variable in this study is customer engagement. Customer engagement on Facebook can be measured in popularity (likes), virality (shares) and commitment (comments) (Bonsón & Ratkai, 2013). The values for likes, shares and comments are collected automatically through the web scraper using the Facebook API. Liking a post is seen as passive participation, where sharing or commenting on a post is viewed as active participation (Barger & Labrecque, 2013).

(33)

Customer engagement – Valence

After obtaining all the data on the per post level, data on the comments of each post need to be collected. This is also done using RStudio, with the same Facebook API that is used for collecting the posts. The objective of this data is to determine the sentiment of the comments to a specific post. This might shed a light on whether customers engage in a positive or negative manner to the post in question. The sentiment analysis is executed by LIWC2015, a computerised text analysis tool used intensively in marketing practices. LIWC2015 can analyse a large bulk of data quickly, where it computes both a score for the positivity of a piece of text as well as a negativity score. Moreover, the tool includes both an English and Dutch dictionary, which is convenient given the nature of the data (Facebook pages were either in English or in Dutch). Because it is too labour intensive to detect the spoken language in each comment, the sentiment of all comments is computed using both the English and Dutch dictionary. Consequently, the sentiment with the highest affect, which is the absolute sum of sentiment positive and negative sentiment, is selected for each comment. The valence for a post is the average of all positive scores minus the average of all negative scores of that post. A post has a positive valence when the positivity score is higher than the negativity score and vice versa. The valence of the post is only neutral when both positivity and negativity scores are equal. An overview of the variables necessary for the analysis is presented in Table 8.

Table 8: Variables used for analysis

Variable Data collection method Variable type Values

Industry type Manual classification Nominal 1 (Controversial) or 0 (non-controversial)

Brand personality Survey Interval 1 - 7

CSR communication Web scraper Nominal 1 (CSR post) or 0 (no CSR post) CSR fit Survey & Web scraper Nominal 1 (good fit) or 0 (bad fit) Customer Engagement - Volume Web scraper Ratio logarithmic scale Customer Engagement - Valence Web scraper Ratio logarithmic scale

(34)

Results

Descriptive statistics

After scraping the Facebook pages, while adhering to the pre-specified methods, 2,240 posts have been collected in total. Only for Google the end of posts was reached in the initial scrape at 76 posts; reason for this was that this page was only initiated on 13th July 2016. After hand coding these initial 1,276 posts, five companies complied to the requirement of 20 CSR posts, and two were found to have no CSR posts at all. A subsequent scrape was executed for the remaining five companies that have shown to post at least some CSR-related messages, but did not pass the threshold yet. For Nike, the end of posts was reached at 260 posts, there other four each completed the limit of 300 posts per company. For some reason, the scraper collected 302 posts for two companies, but this is nothing more than a beauty error and does not affect the consequent analysis. Table 9 contains detailed information about the posts obtained for each company.

Of these 2,240 posts that were collected and hand coded for CSR communication and types, 136 posts did not contain a message. These posts were mainly changes in profile pictures or status updates, which are not of importance to this study, and are thus eliminated from the dataset. Regarding the types of post mainly used by companies, most of the posts contained photos (43%), followed by videos (33%) and links (23%). Status updates (1%) and events (0%) have rarely been used. Of the 328 posts that contain CSR communication, most were concerning the environment (42%), followed by the community support (21%), diversity (18%), human rights (9%) and employee support domain (6%). 4% of the CSR posts could not be classified according to the proposed definitions of CSR types.

(35)

Table 9: Collected Facebook posts per company

When establishing the interrater reliability of CSR communication, there was almost perfect agreement between the two independent raters, κ = .834 (95% CI, .778 to .890), p < .01. This entails that the classification for the variable CSR communication has been found to be correct, and will be used for further analysis. Furthermore, regarding the types of CSR, an almost perfect agreement between the two independent raters has also been reached, κ = .863 (95% CI, .841 to .885), p < .01, which indicates that this classification is suitable for consequent analysis as well.

Customer engagement – Volume

The number of likes, comments and shares do not follow a normal distribution (with skewness and kurtosis statistics greatly beyond 1, see Table 10), most likely due to the viral nature of social media posts. There are only a few posts that obtain a disproportional share of the customer engagement, which positively skews these variables. The mean number of likes, comments and shares and their variance is heavily influenced by these outliers. However, these cases may prove to be interesting to examine, because it is exactly this virality of posts that companies are mostly aiming for on social media. Therefore, instead of deleting these outliers, the logarithmic scale of these values will be used in further analysis. The skewness and kurtosis

# % # % # % # % # % # % # % AB InBev Yes 100 43 43% 1 2% 16 37% 1 2% 25 58% AF KLM No 302 7 2% 2 29% 1 14% 1 14% 3 43% BP Yes 302 23 8% 3 13% 11 48% 3 13% 4 17% 2 9% Coca-Cola No 100 0 ExxonMobil Yes 100 32 32% 2 6% 21 66% 4 13% 3 9% 2 6% Google No 76 1 1% 1 100% Heineken Yes 100 0 ING No 300 27 9% 1 4% 14 52% 1 4% 5 19% 3 11% 3 11% Nike No 260 10 4% 4 40% 2 20% 4 40%

Philip Morris Yes 100 56 56% 10 18% 3 5% 30 54% 9 16% 1 2% 3 5%

Shell Yes 100 46 46% 2 4% 27 59% 3 7% 11 24% 2 4% 1 2% Toyota No 300 23 8% 16 70% 4 17% 3 13% Unilever No 100 60 60% 27 45% 8 13% 6 10% 19 32% Total 2240 328 15% 19 6% 138 42% 59 18% 69 21% 30 9% 13 4% Diversity Community Human rights Other Company Contr. Ind. N of posts

(36)

statistics of these logarithmic scales all lie perfectly between -1 and 1 (see Table 10), which are found to be the acceptable limits for a normal distribution. The log of zero does not exist, therefore the number of observations where the like, comment and share count equals zero are eliminated from their respective logarithmic scale.

Table 10: Descriptives of popularity (likes), commitment (comments) and virality (shares)

Customer engagement – Valence

For the 2,104 posts, a total number of 202,460 comments are collected, which amount to a mean amount of above 96 comments per post, keeping in mind that this amount is positively skewed due to the potential viral behaviour of posts. The lowest sentiment that is captured is -34.00, while the highest sentiment captured is 100.00, with a mean of 7.58 (Table 11). The sentiment is positively skewed, due to a higher amount of positive comments compared to negative comments, and a big bulk of neutral comments. Therefore, the logarithmic scale of sentiment will be used for further analysis to ensure that it follows a normal distribution (Table 11). The logarithmic scale of a neutral sentiment (sentiment = 0.00) does not exist, therefore 350 posts have been omitted from the logarithmic scale for sentiment.

Table 11: Descriptives of sentiment (valence)

Variable N Min. Max. M SD Statistic Std. Error Statistic Std. Error

Likes 2,104 0 411,530 4,510.58 19,751.95 12.35 .05 204.80 .11 Comments 2,104 0 25,143 157.98 730.79 21.56 .05 670.31 .11 Shares 2,104 0 130,273 552.85 4,529.06 18.83 .05 440.35 .11 Log_likes 2,087 .00 5.61 2.64 .90 .38 .05 .19 .11 Log_comments 1,995 .00 4.40 1.37 .83 .42 .06 -.36 .11 Log_shares 1,904 .00 5.11 1.62 .85 .56 .06 .54 .11 Skewness Kurtosis

Sentiment 1,946 -34.00 100.00 7.58 12.23 3.77 .06 21.49 .11

Log_sentiment 1,596 -2.48 2.00 .73 .49 -.46 .06 1.68 .12

(37)

Correlations

Almost all variables are significantly correlated with each other (see Table 12).

A few high correlations indicate that there might exist multicollinearity in the data. Since all companies in the sample have one single score for sincerity, competence and industry type, these are highly correlated. CSR fit is also moderately to highly correlated with these three variables, since CSR fit is also a company-dependent statistic. To test for multicollinearity, the variance inflation factor (VIF) is analysed for these variables. There is evidence of harmful multicollinearity when the VIF is greater than 10 (Dormann et al., 2013). The VIF for multiple variables have been computed in different models, to properly test for multicollinearity when these variables are used together in one model. As illustrated in Table 13, the data does not show evidence of multicollinearity, since the VIF of each variable is well below 10.

The likes, shares and comments variables are all highly correlated as well. Therefore, these variables are computed into one customer engagement volume variable, to avoid redundancy by having do a triple analysis for popularity, virality and commitment. The following formula has been used: Customer engagement volume statistic = likes (log) + Table 12: Means, standard deviation, and correlations

Variables M SD 1 2 3 4 5 6 7 8 9 10 11 12 1. Company 6.82 3.80 -2. Industry type .37 .48 -.30** -3. Post type 1.82 .81 -.17** -.21** -4. CSR communication .16 .36 .17** .22** -.05* -5. CSR type 6.37 1.54 -.16** -.22** .05* -.95** -6. CSR fit (%) .00 15.44 .16** -.68** -.14* .c .27** -7. Sincerity score (%) _.00 _{13.91 .19}** -.76** .11** -.26** .26** .74** - (.86) 8. Competence score (%) _.00 _{7.72 .28}** -.72** .11** -.22** ,23** .61** .95** - (.79) 9. Likes (log) _2.64 _{.90 .22}** .03 -.29** .12** -.12** .18** .15** .10** -10. Comments (log) 1.37 .83 .11** -.19** -.05** -.09** .09** .17** .39** .38** .73** -11. Shares (log) 1.62 .85 .19** -.04 -.16** -.02 .00 .13* .24** .21** .81** .77** -12. Sentiment (log) .73 .49 .09** .09** -.18** .14** -.12** .11 -.14** -.15** .11** -.21** .02 -**

. Correlation is significant at the 0.01 level (2-tailed).

*

. Correlation is significant at the 0.05 level (2-tailed).

c

(38)

comments (log) + shares (log). The logarithmic scales have been used to avoid biases, since the number of likes can greatly surpass the number of shares or comments. As Table 14 indicates, the customer engagement variable is normally distributed with a skewness statistic of .591 and a kurtosis statistic of .066.

Table 13: Variance inflation factors (VIF’s)

Table 14: Descriptives of customer engagement statistic

Analysis

To be able to properly analyse the result, the selected models are being evaluated on customer engagement volume and valence separately. For both volume and valence, the mediation effect of brand personality is tested first (see Figure 2).

Figure 2: Mediation model

Subsequently, the moderation effect of industry type is being introduced in the model (see Figure 3). The two models are tested for both the independent variables CSR communication and CSR fit, to be able to evaluate the differences between these models. The

1 2 3 4

Variables VIF VIF VIF VIF

Company 1.168 1.186 2.419 1.816

CSR communication 1.139 1.137

CSR fit 2.810 2.560

Brand personality - Sincerity 2.367 5.390

Brand personality - Competence 2.108 2.914 Industry type 2.487 2.130 5.353 3.543

Model

Customer_engagement 1,818 .00 14.38 5.806 2.334 .591 .057 .066 .115 Skewness Kurtosis

(39)

categorical company variable is used as a control variable in all the models to control for any company effects that are present in the dataset. All tested models use heteroscedasticity consistent SEs, which prevents heteroscedasticity to influence the models.

Figure 3: Moderated mediation model

Customer engagement – Volume

First, the effect of CSR communication on the volume of customer engagement is tested, both directly and indirectly through the perceptions of brand personality. This test has been performed twice because of the two different brand personalities that are tested: sincerity and competence. Consequently, the effect of CSR fit on customer engagement volume is tested, also for both brand personalities. This is done to be able to compare the predictors CSR communication and CSR fit in the model. All models control for the company effect. For convenience, the volume of customer engagement is addressed with just ‘customer engagement’ in this chapter.

CSR communication and sincerity

Two posts that differ by one unit of CSR communication (so, CSR message or not) are estimated to differ by a1 = -.650 (p < .001) units in sincerity (see Table 15). A CSR post has a

(40)

negative influence on the perception of sincerity. Two similar posts that differ by one unit in sincerity are estimated to differ by b1 = .795 (p < .001) units in customer engagement. Firms

higher in sincerity are estimated to have a higher customer engagement. The indirect effect of a1b1 = -.517 means that CSR communication leads to a decrease in customer engagement,

because of the decreased perceived level of sincerity of a firm (see Table 16). This indirect effect is statistically different from zero, as revealed by a 95% confidence interval that is entirely below zero (-.639 to -.408). There is no significant direct effect of CSR communication on customer engagement (p = .074). The total effect of CSR communication on customer engagement is c1 = -.278, meaning that CSR communications are estimated to differ by .278

units of customer engagement. The negative sign indicates that CSR communication leads to a decrease in customer engagement. This effect is statistically different from zero (p < .05). Thus, customer engagement is negatively affected by CSR communication, but only through the mediating effect of sincerity. The R squared quantifies the proportion of the total variance explained by this model. This model explains 10.80% of the variance in customer engagement, and this is statistically significant (p < .001).

Table 15: Mediation model CSR communication and sincerity

Table 16: Matching total effects model

Antecedent Coeff. SE p Coeff. SE p

CSR comm. (X) a1 -.650 .054 <.001 c1' .239 .134 .074

Sincerity (M) - - - b1 .795 .058 <.001

constant i1 3.957 .043 <.001 i2 1.858 .217 <.001

Consequent

Sincerity (M) Customer Engagement (Y)

R2 = .131 R2 = .108

F(2, 1815) = 124.462, p<.001 F(3, 1814) = 108.701, p<.001

Effect SE p LLCI ULCI

Direct effect c1' .239 .134 .074 -.024 .502

Total effect c1 -.278 .130 <.05 -.533 -.022

Boot SE Boot LLCI Boot ULCI Indirect effect a1b1 -.517 .060 -.639 -.408

The pains and gains of showing off CSR activities on social media