Doing Good in Times of Crisis: The Impact of CSR
Perceptions on Customer Satisfaction and Firm
Recommendations
Based on data from the COVID-19 crisis in the Netherlands
Master Thesis
MSc. Marketing Intelligence & Marketing Management
Faculty of Economics and Business
University of Groningen
Kelly Felicia Schrama
S2711478
Wipstraat 7a
9712 LW, Groningen
The Netherlands
+31 (6) 34 88 51 03
k.f.schrama@student.rug.nl
Supervisor: Dr. Evert de Haan
Second supervisor: Dr. Peter S. van Eck
ABSTRACT
Corporate Social Responsibility (CSR), which is also referred to as “actions that appear to further some social good, beyond the interests of the firm and that which is required by law” (Mc Williams & Siegel, 2001, p.117) are becoming increasingly integrated into firm strategy. Earlier research has highlighted the positive effects that CSR practices may have on customer satisfaction and consumer (loyalty) behavior, including recommendations. The relationship between CSR and customer satisfaction has been researched in the setting of a financial crisis, where it appeared that in times of a crisis consumers increasingly value firm-initiated CSR practices. The current COVID-19 crisis provides new grounds to research the impact of a global crisis, because it has had such a major global impact, both financially and in terms of health. This study uses customer satisfaction data from a panel that contains approximately 89,000 observations, from 19598 different respondents for 343 different firms. To measure the effect of a crisis, crisis data regarding the number of hospitalizations, online search volume, the expected economic impact and strictness of the government measures was gathered from various sources. Multi-level modeling was used to test the formulated hypotheses. I find that perceived CSR in firms positively impacts customer satisfaction and that more satisfied customers are more likely to recommend a firm. Furthermore, I find that a crisis negatively impacts the relationship between perceived CSR in firms and customer satisfaction, but also that the depth of a crisis positively impacts this relationship. Finally, during a crisis it appears that CSR is especially valued in firms that are considered to be essential in times of a crisis. Ultimately, these findings add to the current body of literature on CSR, customer satisfaction, consumer behavior and how a crisis may impact the relationship between these constructs.
PREFACE
Right in front of you lies the thesis “Doing Good in Times of Crisis: The impact of CSR Perceptions Customer Satisfaction and Firm Recommendations”. Writing this thesis was one of the final steps required in order to complete my masters’ in Marketing Management and Marketing Intelligence at the University of Groningen. I was engaged in researching and writing this thesis from September 2020 up until January 2021.
Through writing this thesis I have greatly improved my knowledge regarding the topic of Corporate Social Responsibility, the antecedents of customer satisfaction, consumer behavior, and how the three of them are related to each other. But most of all, how a global crisis can impact these constructs. Writing this thesis during the crisis upon which this thesis was based, the COVID-19 crisis, makes it even more special.
I must admit that writing this thesis has been difficult at times. Therefore, I want to take this opportunity to thank a few people without whom writing this thesis had not been possible. First of all, I want to thank my supervisor Evert de Haan, who has always been very willing to answer my questions and brainstorm with me throughout the entirety of this process. Moreover, I really want to thank Lisette, my family and my friends for the ongoing support throughout this process, it is truly one of the only things that kept me sane. Finally, I want to thank my friends and fellow master students Annebeth and Sophie, who more than anyone have truly understood the stress, effort, issues, problems and what not when writing this thesis. They never hesitated to help me out and cheer me up when I felt overwhelmed at times.
Table of content
1 INTRODUCTION ... 6
2 CONCEPTUAL FRAMEWORK AND HYPOTHESES ... 9
2.1 CORPORATE SOCIAL RESPONSIBILITY ... 9
2.1.1 Drivers of Corporate Social Responsibility... 10
2.2 CUSTOMER SATISFACTION AND CSR ... 11
2.3 CUSTOMER SATISFACTION AND RECOMMENDATIONS ... 12
2.4 THE MODERATING ROLE OF A GLOBAL CRISIS ... 14
2.5 THE MODERATING ROLE OF CRISIS RELEVANCE ... 17
2.5.1 Belonging to a risk group and living in a certain region ... 17
2.5.2 Industries ... 19
2.6 CONCEPTUAL MODEL AND DEVELOPED HYPOTHESES... 20
3 DATA ... 21
3.1 INDIVIDUAL CSR PERCEPTIONS,CUSTOMER SATISFACTION AND RECOMMENDATIONS ... 22
3.2 CRISIS DATA ... 23
3.2.1 RIVM data... 24
3.2.2 Lockdown data ... 24
3.2.3 Expected economic impact data ... 25
3.2.4 Google Trends data ... 26
3.3 DESCRIPTIVE STATISTICS AND MODEL FREE EVIDENCE ... 26
4 METHODOLOGY... 32
4.1 MULTILEVEL REGRESSION ANALYSIS ... 32
4.2 DATA PREPARATION ... 33
4.3 MODEL DEVELOPMENT... 34
4.4 MULTILEVEL MODEL ASSUMPTIONS ... 37
4.5 MODEL FIT ... 39
5 RESULTS ... 39
5.1 IMPACT OF CSR PERCEPTION AND CRISIS ON CUSTOMER SATISFACTION... 39
5.1.1 Effect of CSR and crisis on customer satisfaction in 2018, 2019 and 2020 ... 39
5.1.2 Results of the multilevel panel regression on Customer satisfaction ... 42
5.2 RESULTS OF THE MULTILEVEL PANEL REGRESSION ON POSITIVE RECOMMENDATIONS ... 46
5.3 ROBUSTNESS CHECKS ... 47
6 DISCUSSION AND MANAGERIAL IMPLICATIONS ... 48
7 LIMITATIONS AND FURTHER RESEARCH ... 52
APPENDICES ... 70
APPENDIX 1:LIST OF ESSENTIAL INDUSTRIES ... 70
APPENDIX 2:THE 7 GOLDEN RULES DEFINED BY MARKET RESPONSE ... 71
APPENDIX 3:DISTRIBUTION OF INDUSTRIES IN THE SAMPLE ... 71
APPENDIX 4:GOOGLE TRENDS SEARCH TERMS... 72
APPENDIX 5:HISTOGRAMS WITH DISTRIBUTIONS OF SATISFACTION WITHIN INDUSTRIES ... 72
APPENDIX 6:BOXPLOTS OF THE SATISFACTION DISTRIBUTION WITHIN PROVINCES... 73
APPENDIX 7:VIF STATISTICS OF MODEL INCLUDING QUARTERS AND ALL CRISIS VARIABLES ... 74
APPENDIX 8:VIF STATISTICS OF MODEL INCLUDING ALL CRISIS VARIABLES ... 74
APPENDIX 9:MODEL OUTPUT OF MODELS WITH ALL CRISIS VARIABLES, WITH AND WITHOUT QUARTER VARIABLES ... 75
APPENDIX 10:MODEL SPECIFICATION FOR SATISFACTION AS DEPENDENT VARIABLE ... 76
APPENDIX 11:MODEL SPECIFICATION FOR THREE-WAY INTERACTIONS WITH CRISIS RELEVANCE VARIABLES .. 76
APPENDIX 12:NORMALITY PLOTS ... 77
APPENDIX 13:HOMOSKEDASTICITY PLOT ... 78
APPENDIX 14:OUTPUT OF INTERCEPT-ONLY AND LINEAR REGRESSION MODEL ... 78
APPENDIX 15:AIC,BIC, AND LOG LIKELIHOOD VALUES OF MULTILEVEL, LINEAR, AND INTERCEPT ONLY MODELS ... 79
1
INTRODUCTION
“Creating a strong business and building a better world are not conflicting goals – they are both essential ingredients for long-term success” – William Clay Ford Jr.
Nowadays, Corporate Social Responsibility (CSR) practices are becoming increasingly integrated into firms’ strategy, going hand in hand with increased financial investments. Not only have many large firms made it a habit to publish annual CSR reports (over 90%), but CSR is also gaining increased attention from the financial sector as investors take Environmental, Social, and Governance (ESG) information into account when making major investment decisions (McPherson, 2020). Where a few decades ago, just a handful of firms were actively engaging and investing in CSR, nowadays it is pointing towards doing so is becoming the new normal (Bernow, Klempner, & Magnin, 2017).
In earlier research , CSR has been defined as “actions that appear to further some social good, beyond the interests of the firm and that which is required by law”(Mc Williams & Siegel, 2001, p.117). Although actively engaging in CSR practices is thus voluntary and not legally required by firms, it is difficult to imagine the current and highly competitive business arena without it. Besides offering firms with long-term financial or market related benefits (Luo & Bhattacharya, 2006) and benefits to the consumer and the issue or cause (C. B. Bhattacharya & Sen, 2004), CSR provides businesses the opportunity to differentiate themselves from competition (Daub & Ergenzinger, 2005) and has the capacity to shape and influence the consumer perception relative to firms and their products (Castaldo, Perrini, Misani, & Tencati, 2009; Chaudary, Zahid, Shahid, Khan, & Azar, 2016; Lii & Lee, 2012; van Doorn, Verhoef, & Risselada, 2020). Customer satisfaction in turn, has often been pointed out as a determinant of (future) firm performance (Fornell, Mithas, Morgeson, & Krishnan, 2006; Morgan & Rego, 2006) and behavioral intentions (i.e. recommendations)(Cronin, Brady, & Hult, 2000), emphasizing the need for a better understanding of the forces that drive CSR effectiveness, its effect on customer satisfaction and how it can impact consumer (loyalty) behavior.
a financial crisis, of which has been shown that it may impact the effectiveness of CSR practices, as well as perceived product quality (A. Bhattacharya, Good, & Sardashti, 2020; Giannarakis & Theotokas, 2011). Although firms naturally tend to cut back on CSR spending, this has been proven not to be the best strategy in times of financial distress, as investing in CSR could actually be beneficial to a firms’ corporate social performance and market value (Kashmiri & Mahajan, 2014).
Because financial crises are re-occurring due to fluctuations in the business cycle, much research has been dedicated to investigating its effects on firms in general, but also more specifically on CSR practices and firm and consumer outcomes (A. Bhattacharya, Good, & Sardashti, 2020; Green & Peloza, 2011). However, a crisis, referred to as shared stress as a result of a failure in a system, is multifaceted (Coombs, 2014; Lerbinger, 2012). Global health crises, often the result of pandemics, are not new to the world, ranging from the Spanish flue in the early 1900s, the HIV/AIDS crisis in the 1980’s, SARS in the early 2000’s and Ebola from 2013-2016 (LePan, 2020). Since the beginning of 2020, we have been battling the largest global health crisis in a century (Murray & Lauerman, 2020), as a result of the spread of the COVID-19 virus, which has impacted both the world economy and global health and has already resulted in nearly two million deaths globally (Johns Hopkins University Center for Systems Science and Engineering, 2020). Being a so-called ‘black swan’ event, which is referred to as a highly unlikely event that no one anticipated (Taleb, 2007), the sudden emergence of the COVID-19 crisis has had immense consequences that have changed the world as we know it.
Although global health crises appear to be a historic recurrence, the sole focus on financial crises in the body of literature on CSR prevails, with no research on other types (or combinations) of crises on perceived CSR in firms and customer satisfaction. Furthermore, researchers have pointed out that in the current body of literature on recessions, the influence of personal characteristics on CSR effectiveness has not yet been investigated, leaving room for further research (A. Bhattacharya, Good, & Sardashti, 2020).
specifically investigated the relationship between CSR and customer satisfaction in a crisis setting, and how this ultimately impacts customer behavior. Second, no research has yet investigated the effect of CSR activities on customer satisfaction in a broader crisis setting that affects both global health and the world economy. Scientists argue that this pandemic might not be the last one (Gill, 2020; Walsh, 2020) because there are still many new undiscovered infectious diseases (Gorvett, 2018). Therefore, it is helpful to know for managers whether allocating resources to CSR or changing current CSR policies (Kramer, 2020) in times of a crisis that also affects global health, can still have a positive impact on customer satisfaction and firm recommendations. Particularly because managers face increasing pressure to prioritize those investments that benefit the firm most in terms of performance, due to increased financial constraints. Third, the current body of literature on CSR in a crisis setting leaves room for research on how personal characteristics of consumers or the type of industry may influence the impact that CSR can have on customer satisfaction. Lastly, some industries are considered as more essential during a crisis, which may impact how consumers value CSR practices in these firms during a crisis. Hence, if marketing managers are aware of which consumers they should target during a crisis and if CSR is more or less valued in their industry, they can yield larger returns from their CSR efforts which could ultimately benefit the firm in the long run. Considering this gap and the rationale behind it, the following research question has been formulated:
“What is the impact of a global crisis on the relationship between individual customer perceptions of Corporate Social Responsibility in firms and customer satisfaction, and ultimately recommendations?”
indicating that during a crisis consumers value CSR practices to a lesser extent. However, I also find that the depth of a crisis positively impacts this relationship, indicating that the more severe the crisis, the more consumers value CSR in evaluating a firm. Because the effects of CSR on customer satisfaction already slightly diminished in 2019, it is to be further investigated whether the decreasing impact of CSR in 2020 is to be attributed to the crisis, or a general trend that CSR is becoming less valued by consumers. Finally, during a crisis it appears that CSR is especially valued in firms that are considered to be essential in times of a crisis. Ultimately, these findings add to the current body of literature on CSR, customer satisfaction, consumer behavior and how a crisis impacts the relationship between these constructs.
2
CONCEPTUAL FRAMEWORK AND HYPOTHESES
This chapter discusses the theoretical framework behind this study. It will discuss and evaluate the existing literature of the research streams of Corporate Social Responsibility (CSR), customer satisfaction, recommendations, crises, and how they are or can be related. Thereafter, the hypotheses will be developed and visualized in a conceptual model.
2.1
Corporate social responsibility
In the field of marketing, Corporate Social Responsibility (CSR) has been a widely studied and prevalent topic for a couple of decades. Nevertheless, it appears to be difficult to reach a consensus on how CSR can be defined. A recent definition has been “actions that appear to further some social good, beyond the interests of the firm and that which is required by law”(Mc Williams & Siegel, 2001, p.117). Moreover, CSR is often explained by the ‘Triple Bottom Line’ (TBL), also referred to as ‘People, Planet, Profit’, which entails firms aligning their efforts with the three pillars of sustainability: social responsibility, environmental responsibility and economic responsibility (Elkington, 1997).
In more recent literature, the concept of CSR has increasingly been approached from stakeholder or business ethics theory (Caroll, 1999). This wide array of literature combined with a lack of consistency in how CSR is defined has sometimes been the cause of confusion (McWilliams, Siegel, & Wright, 2006). Namely, the concept of CSR is also intertwined with various other related constructs (Caroll, 1999), including Corporate Sustainability (van Marrewijk, 2002), Corporate Citizenship (Wood & Logsdon, 2002) and Corporate Social Performance (Wood, 1991). Despite the lack of congruency among formal definitions of CSR, a study by Dahlsrud (2008) showed that all formal definitions of CSR contain elements of the following five dimensions: environmental, social, economic, stakeholder, voluntariness.
2.1.1 Drivers of Corporate Social Responsibility
Some authors argue that firms engage in CSR simply to cover their wrongdoing (Freeman, 2010) or that it could have a negative impact on financial performance (Wright & Ferris, 1997). However, throughout the field of literature of CSR, it is known to benefit firms in many ways. First of all, firms can benefit from a well-known and positive CSR track record or a public image that is often associated with CSR, because in the current era, social goals are gaining more priority amongst consumers (Davis, 1973). Because CSR practices can positively influence how consumers view a firm (Sen & Bhattacharya, 2001), when making a purchase decision, consumers can take into account their perception of CSR performance of a firm (Mohr, Webb, & Harris, 2001). This is especially the case for products that symbolize ethical or social values (Castaldo et al., 2009). Besides the positive effect that CSR practices have on perceived brand quality by consumers (Mc Williams & Siegel, 2001), van Doorn, Verhoef & Risselada (2020) showed that whereas making a sustainability claim for a product may initially harm the perceived quality of a product, this negative association can be reduced for firms that actively participate in CSR and have a good CSR track record.
elements for all consumers, they might still identify themselves with a firm or brand based on their positive CSR track record and/or reputation (Lii & Lee, 2012). Hence, if consumers make purchases that support corporate philanthropy, this can aid consumers in enhancing their self-esteem, which is beneficial for both consumers and firms (Basil & Weber, 2006; Ramasamy, Yeung, & Au, 2010).
Additionally, excelling in CSR activities is considered to be an important differentiation strategy, because in order for a brand to be noticeable for consumers, it should stand out and be different (Kay, 2006). By doing so, firms are able to differentiate its products or brands from that of competitors (Smith & Higgins, 2000), ask for a price premium or create new demand (Mc Williams & Siegel, 2001). Moreover, perceived differentiation by consumers is argued to be one of the key drivers of customer loyalty (Bennett & Rundel-Thiele, 2005).
Finally, CSR has been coined as a driver of several financial firm performance metrics. Although authors argue that it is difficult to link CSR to increased firm performance directly, research indicates that CSR is able to influence firm performance indirectly, through several mediators such as customer satisfaction, reputation and competitive advantage (Luo & Bhattacharya, 2006; Saeidi et al., 2015).
2.2
Customer satisfaction and CSR
Customer satisfaction describes the judgement or evaluation of a consumer whether a good or service has lived up to a consumers’ expectations of how it should perform over time (Gupta & Zeithaml, 2006; Luo & Bhattacharya, 2006), in line with the confirmation/disconfirmation framework proposed by Yi (1989). According to Gupta & Zeithaml (2006), customer satisfaction is an unobservable or perceptual customer metric, that tries to measure what consumers think after certain actions taken by a firm (i.e. marketing actions).
Anderson, Cha, & Bryant, 1996). Numerous studies that investigate customer satisfaction build upon these measures, and research points towards the notion that satisfied customers will result in many benefits for firms, both in terms of behavioral outcomes and financial performance (Gupta & Zeithaml, 2006).
Customer satisfaction and CSR are two concepts that have also been studied together. In line with Daub & Ergenzinger's conceptualized 'generalized customer' (2005), some consumers ought to be increasingly satisfied by products of firms that are very involved with CSR. When firms engage in CSR, this has a positive influence on the perceived utilitarian, emotional, and social value of customers (Currás-Pérez, Dolz-Dolz, Miquel-Romero, & Sánchez-García, 2018). Other studies have pointed out that in some industries, a higher perceived value can positively influence customer satisfaction (H. H. Hu, Kandampully, & Juwaheer, 2009; Kuo, Wu, & Deng, 2009; McDougall & Levesque, 2000; Yang & Peterson, 2004). Hence, one could expect that CSR must also be positively related to customer satisfaction.
Some studies have specifically investigated the direct relation between incorporating CSR in a firms’ strategy and customer satisfaction. For example, if a firm is known for actively engaging in CSR practices, this creates a certain favorable context for customers that may improve their evaluations and attitudes towards a firm (Mandhachitara & Poolthong, 2011; Sen & Bhattacharya, 2001). However, this applies to firms that are known to be innovative and have high product quality. Nevertheless, a later study Luo & Bhattacharya (2006) found that CSR practices in firms have a direct and positive influence on customer satisfaction. The same relationship was found in similar studies conducted about CSR in China, foreign MNCs Malaysia and the shipping industry in South Korea (Chung, Yu, Choi, & Shin, 2015; Hassan & Nareeman, 2013; Shin & Thai, 2015). In line with these findings, the following hypothesis is also expected to apply to this study:
2.3
Customer Satisfaction and Recommendations
How consumers behave and especially what drives their behavior has been a topic that marketeers have been intrigued by for decades. In order to be able to predict future customer behavior, firms often try to make use of metrics, which are referred to as Customer Feedback
H1: The individual customers’ perception of CSR in firms is positively related to customer
Metrics (CFMs: Morgan & Rego, 2006) that say something about future purchase behavior or future firm performance.
The Net Promoter Score (NPS) is a type of CFM and is a measure developed by Reichheld (2004), that calculates the ratio of people who are likely to recommend a company (‘promoters’) to the number of people that would not (‘detractors’) in a scale of -100 to +100. While Reichheld (2004) argues that NPS is the most important metric for predicting future growth, there has been quite some controversy about this metric in the literature. A study by Morgan & Rego (2006) found that general customer satisfaction has been shown to be the best predictor of future firm performance, and that net promoters (a proxy for NPS) and number of positive recommendations have little to no predictive value. However, a study by van Doorn, Leeflang, & Tijs (2013), that replicated the study of Morgan and Rego (2006) for Dutch firms, found that NPS was equally good at predicting firm performance as other customer feedback metrics. Moreover, de Haan, Verhoef, & Wiesel (2015) found that NPS has an impact on retention, and is still a good metric in order to identify “top churners”, although there is much heterogeneity among industries.
2.4
The moderating role of a global crisis
The setting in which this study is conducted, involves a large-scale global crisis as a result of a pandemic that so far has not only impacted the global economy, but also the health of in inhabitants all over the world. Because it is the first time in recent history that a crisis like this has occurred on such a large scale, naturally no research has been conducted on the relationship between CSR and customer satisfaction in a similar setting. Hence, hypotheses will be formulated based on knowledge of prior crises and consumer behavior.
In order to establish the antecedents of consumer behavior in times of crisis, a basic understanding of what drives consumer decision making is required. Considering Maslow’s hierarchy of needs, people have to satisfy their basic needs (physiological and safety) before they can be concerned about psychological and self-fulfillment needs (Saul Mcleod, 2007). Consumers pay attention to CSR in firms for various reasons, for example because it resonates with their personal values and provides them with a positive self-concept. Hence, if a consumers’ basic needs are not met in the midst of a global crisis, there is little room for consumers to fulfill such esteem and self-actualization values. However, a study on the influence of CSR on purchasing, could not confirm that in a more developed country, where basic needs are satisfied to a larger extent, CSR was valued more (Marquina & Morales, 2012). In order to create a better understanding of how consumers may behave during a global crisis, I will focus on earlier research that was conducted on CSR in the setting of financial crises
When focusing on financial crises only, in the history of the past century, two major global recessions have occurred that affected the entire global economy (Investopedia, 2020b; Walsh, 2020). The first was the Great Depression, which started in October 1929 after the stock market crashed and lasted until 1939 (History.com, 2020). The second recession, which is often referred to as the Great Recession, occurred from 2007-2009 and was the result of the bursting of the US housing bubble (Investopedia, 2020a). Due to the large financial losses that firms suffer from during such financial crises, firms often cut back on R&D, marketing, rates of New Product Introduction (NPI) and CSR related expenses in order to control their finances (Bansal, Jiang, & Jung, 2015; Kashmiri & Mahajan, 2014; Srinivasan, Lilien, & Sridhar, 2011). However, cutting back is not always the best strategy, as it could in some cases be detrimental
to a firms’ long-term competitive position in the market (Srinivasan et al., 2011). Part of this can be attributed to the fact that consumers become more price sensitive during crises. While consumers respond less to advertising during economic downturns, advertising does act as a buffer against increased price sensitivity because it helps to determine consumer preference (Gijsenberg, 2013; Van Heerde, Gijsenberg, Dekimpe, & Steenkamp, 2013).
Still, during a financial crisis, CSR is often seen as a threat to business survival because of its associated costs (Souto, 2009). Nevertheless, if companies improve or maintain their CSR performance this could aid in regaining the lost trust in the company, transforming this threat into an opportunity. This was shown by a study of Kashmiri & Mahajan (2014), that found that family firms outperform non-family firms during a recession, because they emphasize CSR more strongly. They argue that during recessions, marked by challenging financial conditions and corporate mistrust, customers may increasingly penalize companies that behave unethically during such a crisis. Hence, engaging in CSR could be very beneficial in times of crisis.
However, some literature contrasts this view, by stating that financially constrained consumers are less likely to make decisions that benefit societal interests, as they are more concerned about their own needs (Flatters & Willmott, 2005; Green & Peloza, 2011). Nevertheless, more recent research revealed the opposite, showing that firms engaging in CSR initiatives during recessions experience an increase in their brand perceptions and value, because it signals higher brand quality to customers (A. Bhattacharya, Good, & Sardashti, 2020). If a consumer thus perceives a firm to actively engage in CSR during a crisis this positively impacts a customer’s evaluation of a company or brand. Another recent study by the same authors confirms this and shows that during recessions consumers are more aware of risks associated with a purchase and thus search for more information. This results in CSR becoming more salient, increasing the perceived quality for consumers (A. Bhattacharya, Good, Sardashti, & Peloza, 2020).
and increased liking for brands that combine functional performance with prosocial behavior (He & Harris, 2020; Knowles, Ettenson, Lynch, & Dollens, 2020).
Second, in times of crisis, consumers have come to realize the importance to helping others in need, with an increased focus on family and community, but also the self. Additionally, during the crisis, moral and ethical values have been receiving greater attention (Euromonitor, 2020a, 2020b). People are moral agents by nature that care about other people and want to prevent themselves from appearing to be selfish (Aquino & Reed, 2002; Nowak & Sigmund, 2005). Consumers do not only judge their own behavior but also that of others, based on the extent to which they behave in a prosocial manner (He, Li, & Harris, 2012). Because firms are increasingly expected to aid in battling the worlds’ past and current issues (Frynas, 2009) and society as a whole has to battle the current COVID-19 pandemic, this could raise consumers’ expectations of firms engaging in CSR practices (He & Harris, 2020). Therefore, when firms start initiatives during a crisis that can help both firms and civilians (De Ondernemer, 2020), and adapt their goals to contribute to tackling the COVID-19 crisis (Hoekstra & Leeflang, 2020), this will not go unnoticed by consumers. Prior research has shown that the CSR associations that a consumer has with a firm, positively influences their social self-concept connection to the company. Accordingly, the social self-concept connection to the company positively influences corporate brand loyalty (Moon, Lee, & Oh, 2015). When a firm successfully manages to convince their customers that their CSR practices are sincere and not self or strategic-serving, this can create favorable attitudes towards companies and positively impact customer satisfaction (Kim & Lee, 2015). Thus, based on these earlier findings, it is hypothesized that a global crisis will strengthen the relationship between CSR and customer satisfaction. Hereafter, the hypotheses are formulated as follows:
H3a: A crisis moderates the positive relationship between perceived CSR and customer
satisfaction; the positive relationship between perceived CSR and customer satisfaction becomes stronger during a crisis.
H3b: The depth of the crisis positively moderates relationship between perceived CSR and
customer satisfaction during a crisis; the worse the crisis, the stronger the moderator effect of crisis on perceived CSR and customer satisfaction
2.5
The moderating role of crisis relevance
Although a global crisis has overall detrimental consequences, its impact is not evenly spread. For example, some people may be more susceptible to experience the negative effects of a health crisis because they fall within a ‘risk group’ and are thus more likely to get severely ill from a contagious virus. Some regions may experience a larger impact because, in the current case of COVID-19, the pandemic spreads at a faster rate in their region. Finally, some businesses are essential during a crisis and will have to keep operating in order to provide society with basic goods, services or infrastructure. These differences among people, regions, and industries could possibly impact the extent to which consumers value CSR during a crisis. The following section will elaborate upon the rationale behind this and formulate hypotheses on how they are expected to impact this relationship.
2.5.1 Belonging to a risk group and living in a certain region
Due to a large variety of customers, not every single one may be equally influenced by a firm’s participation in CSR. According to Bhattacharya & Sen (2004), there is much heterogeneity among consumers and their responses to CSR, which also becomes evident from contrary findings in literature. Where some studies have not found any significant difference between males and females in their orientation to CSR (Kahreh, Babania, Tive, & Mirmehdi, 2014; Marquina Feldman & Vasquez-Parraga, 2013), another study found that more generally, females tend to value ethical responsibilities to a larger extent than males (Haski-Leventhal, Pournader, & McKinnon, 2017). Age appears to be mildly related to CSR attitudes (Haski-Leventhal et al., 2017), although one study found that older consumers feel increasingly morally responsible for their shopping behavior (Carrigan, Szmigin, & Wright, 2004).
This perceived risk is could be influenced by the availability heuristic, that people often use when they are under high cognitive load, such as a crisis. This type of heuristic is a mental shortcut where people rely, for example, on memory or imaginability to assess a certain risk (Lichtenstein, Slovic, Fischhoff, Layman, & Combs, 1978). In the case of the current COVID-19 pandemic, when people are repeatedly exposed to information that stipulate a risk associated with age, they are thus more likely to use this as a source of information to judge their own perceived risk. This is also in line with the findings of another study, that showed that higher mortality salience lead to increased state anxiety (J. Hu, He, & Zhou, 2020). Therefore, there are some (age) groups in society that feel as if they are at greater risk to experience severe consequences from getting ill during the current COVID-19 crisis.
Furthermore, in the development of the pandemic in the Netherlands, not all regions have been hit equally hard in terms of the number of cases and deaths. Therefore, a customer from Brabant, which initially had many cases and hospitalizations, could have a different perception regarding CSR than a customer from Groningen, which initially experienced one of the lowest numbers of COVID-19 cases. The different impact of the crisis across regions was investigated by a study of Neuteboom, Golec, & Phlippen (2020), which showed that the number of pin transactions in the Netherlands during the first wave was much lower, compared to the same period in the previous year, in a province that had a higher number of cases compared to a province that had much lower cases. They conclude that consumers spend less when the virus spreads at a faster rate, regardless of the lockdown measures. With regards to risk perceptions of people living in more severely hit regions, an early study on risk perception regarding the COVID-19 crisis was found that in regions with more victims of the virus, people were inclined to overestimate the risk of the virus (Abel et al., 2020). People that live in regions with more victims are more likely to know someone who either died from the virus or became very ill from it. Accordingly, by applying the availability heuristic (Butler & Mathews, 1987; Shaham, Singer, & Schaeffer, 1992), they perceive higher risk associated with the virus (Abel et al., 2020). This ultimately decreases analytical or rational thinking and therefore affects risk perception, being consistent with the findings of Keller, Siegrist, & Gutscher (2006).
Ming-Sung Cheng, 2020). A recent study by Schimmenti, Billieux, & Starcevic (2020) showed that a certain threat or risk is fear-arousing. They argue, that in a case such as the COVID-19 crisis, people are expected promote responsible and moral behavior in order to confront this fear. Therefore, if people perceive higher risk, they are increasingly inclined to behave in a more prosocial manner. Using the same reasoning as for Hypothesis 3, it is expected that increased expectations of prosocial behavior among consumers increases their expectations of others and firms to behave similarly. Therefore, it is expected that those people who are more at risk and live in regions that were more severely hit by COVID-19 (and thus perceive higher risk), also place higher value on firms behaving prosocial through CSR practices. Furthermore, the impact of the crisis might have been heaviest for the respondents in risk groups and/or from severely hit regions when it was at its peak. Therefore, it is expected that this strengthen the moderating role of the depth of a crisis on the relationship between perceived CSR and customer satisfaction. The corresponding hypotheses are formulated as follows:
2.5.2 Industries
Whenever a crisis occurs, this influences consumer behavior and how consumers make purchases (Laato, Islam, Farooq, & Dhir, 2020; Sheth, 2020). For example, stockpiling on certain items such as toilet paper, has been argued to be rational consumer behavior during a crisis, where the consumer is faced with a large amount of uncertainty (Lufkin, 2020). Here, it is important to take into consideration how essential a certain industry is during a crisis, in order to satisfy basic (human) needs. The Cambridge University Press (2020) defines an essential industry as: “an industry that is considered necessary for a nations’ economy and that may be
H4a: In a crisis setting, belonging to a risk group strengthens the relationship between perceived
CSR and customer satisfaction
H4b: In a crisis setting, the extent to which the region in which a customer lives was impacted by
the crisis, strengthens the relationship between perceived CSR and customer satisfaction. Here, the effect is strongest for customers living in a severely hit region.
H5a: In a crisis setting, the extent to which the region in which a customer lives was impacted by
the crisis, strengthens the positive moderator effect of the depth of a crisis on the relationship between perceived CSR and customer satisfaction.
H5b: In a crisis setting, the extent to which the region in which a customer lives was impacted by
protected or supported by the government”. However, the Dutch government considers those industries that are needed to keep society running to be essential (Overheid.nl, 2020a). For the full list of essential industries as defined by the Dutch government please refer to Appendix 1.
When people in the Netherlands initially panicked in the very beginning of the crisis, this resulted in extremely busy stores and empty shelves because customers feared that the supermarkets would not be able to restock the stores (Business Insider, 2020). Moreover, now that most people are working from home in order to reduce mobility of commuters (de Haas, Faber, & Hamersma, 2020), people rely more than ever on their internet providers to provide them with stable Wi-Fi connections (Stil, 2020). A recent study showed when store employees violated COVID associated norms, this reduced customer satisfaction (Söderlund, 2020). Therefore, it is expected that for essential firms, consumers will place more value on the firms’ corporate abilities to efficiently and safely provide them with those services that satisfy their basic needs rather than their CSR practices. For non-essential firms, consumers will still be able to judge a firm by how it is engaged in CSR practices to a larger extent, as their goods and services are less critical during this crisis. Furthermore, because the impact of the crisis might have been heaviest for the essential firms when it was at its peak, it is expected that this also weakens the moderating role of the depth of a crisis on the relationship between perceived CSR and customer satisfaction. This results in the following hypotheses:
2.6
Conceptual model and developed hypotheses
This thesis focuses on the relationship between a consumers’ perception of whether a firm engages in CSR, and how and to what extent this relationship is influenced by a global crisis. Although extensive research has been conducted on this main relationship, the influence of a global crisis on this relationship has not yet been studied explicitly. This study therefore aims to address this gap. The conceptual model for this thesis is split up into two parts (a and b) to enhance the ease of interpretation, which are presented in Figure 1 and Figure 2. Figure 1 depicts the main relationship (a) between the individual customer perception of CSR of firms H4c: In a crisis setting, the type of industry weakens the relationship between perceived CSR
and customer satisfaction; the relationship will be negative for industries that are essential during times of crisis
H5c: In a crisis setting, if a firm is essential, this weakens the positive moderator effect of the
and customer satisfaction, and the relationship (b) between customer satisfaction and recommendations. The model includes one moderator (c), namely the effect of a crisis which is expected to positively influence relationship (a). Customer satisfaction thus serves both as a DV and as a mediator in this model.
Figure 1: Conceptual model a
Once the effect of a crisis on relationship (a) has been established, the effects of other moderators can be analyzed in a crisis-only setting, only using data from the period when there was a crisis. This is visualized in Figure 2. The model includes two moderators, namely the effect of the depth of a crisis (d) and crisis relevance (e), which depends on the level of personal threat, region in which a consumer lives and the type of industry in which a firm operates. Each of these moderators is expected to influence relationship (a). In addition, the model includes three-way interactions of the crisis relevance variables with the depth of a crisis and CSR perception. These three-way interactions are expected to influence the moderator effect (d)
Figure 2: Conceptual model b
3
DATA
The data sources will be described in the following subsections. Finally, this chapter will provide some descriptive statistics of the final dataset.
3.1
Individual CSR perceptions, Customer satisfaction and
Recommendations
For the measurement of customers’ individual CSR perceptions of firms, how satisfied they are about the firm, and whether or not they recommended the firm, a dataset was provided by the data provider Market Response. Market Response is a firm that conducts market research among Dutch consumers and firms. They do this by surveying customers about the companies they are a client of, how they rate their customer service, whether the customers have recommended the company to anyone in the past two years and if they think the company sticks to the so-called “seven golden rules” (refer to Appendix 2 for the full list) (Piksen, 2020). Of the seven golden rules, which are measured on a 0/1 scale (1 = respondent thinks the firm fulfills the formulated rule), not all are relevant for this research. Therefore, the rule: “Be concerned about people and society” will be used as a proxy of an individuals’ perception of CSR in the companies. How the customers ranked the service of the company a 1-10 scale will be used as a proxy for customer satisfaction. The data on whether or not a respondent has recommended a firm (1/0 variable, 1 = respondent has recommended the firm) will be used as a measure for the number of recommendations.
The dataset contains data covering the period from the 16th of February 2018 up until the 5th of
Although the dataset contains observations for the period 2018-2020, the number of observations is somewhat lower in the months March and April for all three years. The reason for this is that during these months the data provider evaluates the list of companies and gather little to no response. Nevertheless, from May onwards the data provider gathers more response, which is included again in the dataset.
3.2
Crisis data
In order to measure a crisis, recent data on the ongoing global COVID-19 crisis has been used to measure how the depth of a crisis may impact the relationship between CSR and customer satisfaction. Because the dataset of the data provider contains information on the Dutch firms and customers only, Dutch data on the developments of the COVID-19 crisis will be used to perform the analyses.
It will not come as a surprise that the current pandemic has impacted the Netherlands. Starting in Wuhan, China, the Corona Virus has now affected people all around the world, totaling to an estimated number of over 85 million confirmed cases and nearly 2 million deaths as of January 2021 (Johns Hopkins University Center for Systems Science and Engineering, 2020). In the Netherlands, these numbers are over 800.000 and nearly 7.500 respectively as of January 2021 (Het Parool, 2020). Because there is no script on how to tackle such a pandemic, each country has been able to develop its own measures on how to battle the virus, resulting in a wide variety of measures. Whereas some governments have decided to introduce a complete lockdown, others, including the Netherlands, have decided to introduce a partial, or “intelligent” lockdown where citizens are still allowed to walk around outside without permission (BBC News, 2020).
This global crisis has had a major impact on the global economy, with big negative shifts in the stock market and large increases in unemployment globally (Jones, Palumbo, & Brown, 2020). In the Netherlands alone, according to the first calculations the GDP has decreased with 8.5% in the second quarter compared to the first quarter of 2020, compared to a decrease of 4.6% in the second quarter compared to the first quarter in 2019 (CBS, 2020b).
initially went down, from September onwards this number started to rise again, and at the moment of writing this thesis we find ourselves right in the midst of the second wave. In order to measure the extent to which the crisis impacts the relationship between individual CSR perceptions of firms and customer satisfaction, the depth of the crisis has to be quantified. However, this does not depend only on the number of cases or deaths, but for example also on economic factors and government measures. Therefore, the depth of the crisis will be quantified by means of using a variety of variables from several data sources:
3.2.1 RIVM data
In order to measure the depth of the COVID-19 crisis, this study will consider the number hospitalizations. The reason for only using the number of hospitalizations instead of also looking at the number of cases and the number deaths as a proxy for crisis severity is twofold. First of all, in the early stages of the crisis there was a lack of testing facilities in the Netherlands. Therefore, the number of cases and deaths as a result of the virus that were reported in the beginning of the pandemic is estimated to be under representative of the actual number of cases and deaths in the country (Deloitte, 2020; Nieuwenhuis, 2020). Moreover, the PRC test that is being used in the Netherlands to test whether someone has been contaminated with the virus, is not 100% accurate, adding to the inaccuracy of number of cases throughout the crisis as a whole (Korteweg, 2020; RIVM, 2020b).
Because the Dutch government strives to be transparent about the number of hospitalizations, the National Institute for Public Health and the Environment (RIVM) has published various data sources that keep track of the numbers (Overheid.nl, 2020b). The dataset that will be used is from the RIVM itself and contains information for various variables, i.e. the number of cases, number of hospitalizations and number of deaths for each province and municipality in the Netherlands on a daily level for the 12 provinces . The dataset is updated daily and contains a total of 100,203 observations in the period starting from the 27th of February until the 15th of
November 2020. However, only data up until the 5th of October will be used to match the data
from Market Response.
3.2.2 Lockdown data
where all inhabitants of the Netherlands were expected to stay at home as much as possible, although they were still allowed to move around freely outside. Moreover, restaurants, bars, schools gyms and other public facilities were closed to minimize mobility among inhabitants as a tool to prevent and limit the contagion of the virus (de Haas et al., 2020).
Since then, the level of measures taken have varied quite a bit over the past couple of months, which is indicative of the depth of a crisis as the more severe measures reflect the corresponding increase in the spread of the virus throughout the Netherlands. Here, the government bases its measures on 4 risk levels: “alert”, “worrisome”, “serious” and “very serious” (Rijksoverheid, 2020). In a report by Neuteboom, Golec, & Phlippen (2020) published by ABN Amro, the ‘Stringency Index’ was used, that displays and quantifies the lockdown style policies by governments all over the world (Oxford University, 2020). Therefore, this index will be used as a quantification of the severity of the lockdown.
3.2.3 Expected economic impact data
3.2.4 Google Trends data
Google Trends (https://trends.google.com) is an online platform initiated by Google that illustrates how often a specific search query has occurred on Google in comparison to global search volume, which therefore also can be used as a proxy for popularity (Kristoufek, 2013). It thus allows you to enter a search term, i.e. “COVID-19”, in order to show how often it has been searched for in a certain time period and location. It also computes an index for each point in time that shows its popularity compared to the highest peak, for which the index equals 100 (see Figure 2 as an example of online search behavior for “Covid-19” in the Netherlands).
The reasons for using google trends are threefold: first of all, extracting search behavior data using Google Trends is an uncomplicated and fast procedure, that allows you to extract data for single search terms, but also allows you to compare up to five search terms. However, for the purpose of this study data for individual search terms were extracted. Second, it allows you to specify your search based on a certain location (as long as there is enough data for this specific location) and also a specified time period. Finally, Google Trends allows you to search for an unlimited number of search terms, without having to pay a fee if you exceed a certain number of enquiries. In total, enquiries for 13 different search terms (see Appendix 4) were run.
3.3
Descriptive statistics and model free evidence
Prior to investigating and analyzing the data in order to test the hypotheses, the dataset was prepared by consolidating data from various sources in order to create the necessary variables. This resulted in a dataset with a total of 86109 unique observations of 19598 respondents and 343 firms. Table 1 provides an overview of the descriptive statistics of the variables in the dataset. There is data for the IVs of customer satisfaction and recommendations, namely individual CSR perception of firms and customer satisfaction itself, as well as data for all the crisis related variables. They vary somewhat in terms of the number of observations, as the crisis variables such as hospitalizations, average search volume and stringency index have only
become available since the start of the Corona crisis. The expected economic impact variable has monthly observations for every month in throughout the time period of the dataset. In terms of variation, especially the hospitalizations national variable has a large standard deviation, which can be attributed to the fact that there is a large variety in the severity of number of hospitalizations throughout the corona crisis. On regional basis, the number of hospitalizations was lower when it reached its maximum, with 99 hospitalizations. Because the province where a respondent comes from is known, this can be matched to the hospitalizations at a regional level. Therefore, the regional hospitalizations will be used for the analyses.
Table 1: Descriptive statistics of the DVs and IVs
Figure 2 shows the distribution of customer satisfaction
in the sample, for all three years. It appears that there are quite some outliers taking on the values as low as 1-5 and as high as 10. However, this could be the result of respondents being either extremely satisfied or dissatisfied with a company, which explains the low or high customer satisfaction ranking. Because it is assumed that these respondents deliberately provided such a low or high score and the outliers are thus not a result due to some error, it was decided to keep the outliers.
Although differences in the distribution of customer satisfaction for each year are not visible in separate boxplots, an ANOVA (p=0.000) showed that customer satisfaction was significantly different from one another for each year, where customer satisfaction was highest in 2020 compared to 2019 and 2018. Customer satisfaction also appears to vary throughout the year,
N Mean Median Std. Dev. Min Max
PerceptionCSR 86109 0.37 0.00 0.48 0.00 1.00 Customer satisfaction 86109 7.5 8.00 1.54 0.00 10.00 Recommendations 86109 0.22 0.00 0.41 0.00 1.00 Age 19598 58.56 61.00 13.56 18.00 95.00 Hospitalizations national 222 58.24 10.00 113.51 0.00 695.00 Hospitalizations regional 222 0.7282 0 2.84 0 99
Average search volume 268 17.15 14.58 12.91 0.08 77.58
Stringency index 279 43.37 47.22 28.18 0.00 79.63
Expected economic impact 33 0.01 -1.67 19.47 -33.33 27.67
indicating some sort of seasonality trend, as an ANOVA (p=0.00) pointed out that customer satisfaction was significantly lower towards the end of the year (Quarter 4) compared to earlier in the year. A similar trend holds when zooming in on 2020, although it is important to note that for 2020 there are only observations up until the 5th of October, which add up to a rather
low number of observations for Q4 in 2020.
When comparing the distribution of customer satisfaction within industries as shown in Figure
3 (for histograms please see Appendix 5), it appears that the distribution of customer satisfaction
is similar for airlines, amusement parks, charity, hostel and lodging, online retail, travel agencies and zoos, each having similar median of 8 and rather large interquartile range, indicating quite some variety in customer satisfaction ratings. For insurance, retail, supermarket and telecom companies the median is similar, but the interquartile range is shorter, indicating less variety in customer satisfaction ratings. For energy and public transportation companies the interquartile range is similar to the previous group of industries, but the median is lower. The banking industry does not show similarities with any of the other industries, as its maximum and minimum values (excluding outliers) are the largest, also having the lowest lower quartile out of all industries. This could indicate that out of all industries, the respondents are generally least satisfied with banks.
This is confirmed by an ANOVA, that showed that customer satisfaction among industries is significantly different from one another (p=0.00), and that all industries except for public transportation show significantly higher customer satisfaction compared to banks. Furthermore,
running an ANOVA between essential and non-essential firms points out that throughout the period of 2018-2020, customer satisfaction is somewhat higher (p=0.00) for non-essential firms compared to essential firms (see appendix 5 for histograms). Within 2020, this difference is only marginally significant (p=0.0934) shows a similar result, where customer satisfaction is higher for non-essential firms compared to essential firms.
As for the distribution of customer satisfaction within the twelve provinces (see Appendix 6), there appears not to be much variation. The interquartile range appears to be similar for all provinces, and only in the province of Groningen the median is somewhat lower compared to the other provinces. The results of an ANOVA show that the customer satisfaction differs significantly within the provinces (p=0.00), which is also the case when only considering 2020 (p=0.00). Throughout the whole sample, although not statistically significant between all combinations of province, customer satisfaction in Drenthe is higher compared to all other provinces, whereas it is generally lower in Flevoland, Friesland and Groningen. In 2020, these results appear to be consistent. When the regions are grouped according to how hard they have been hit in terms of no. hospitalizations (worst hit > medium hit > least hit), it appears that in 2020 customer satisfaction is significantly lower in the least hit provinces compared to medium and worst hit regions. However, these differences only appear to be significant in 2020, not throughout the total period.
probably lead to a worsened expected economic impact. At first glance this also appears to be visible in Figure 4. However, search volume appears to spike first, which is probably due to the initial interest in what COVID-19 entails prior to the spike in hospitalizations. Stringency spikes somewhat around the same moment, and except for a drop during summer it remains rather stable throughout the year. There appears to be some correlation, but it is not necessarily expected that this may cause problems in terms of multicollinearity.
When judging the impact of the crisis variables on customer satisfaction, it appears that the
variables ‘StringencyIndex’ and ‘SearchVolume’ are significantly and positively correlated to customer satisfaction, whereas ‘ExpectedEconomicImpact’ is significantly but negatively correlated to customer satisfaction. Only the last crisis variable ‘Hospitalizations’ is not correlated with customer satisfaction. Ultimately, this could be some initial evidence for the hypothesis that the depth of a crisis may (positively) impact customer satisfaction.
By performing some initial regression analyses the expectation that CSR has a positive effect on customer satisfaction is confirmed, as this showed that in each year (2018, 2019 and 2020), CSR has a significant positive impact on customer satisfaction (p<0.001 for all three years). However, this positive effect appears to be smallest in 2020, which could point towards the notion that CSR may become less important during a crisis. Considering that some of the crisis variables appeared to be positively correlated with customer satisfaction, this is something worth to look into further to see if this is actually the case.
Table 2:Correlation matrix of the variables of interest
Customer
Satisfaction PerceptionCSR Recommendation Hospitalizations StringencyIndex
Expected Economic Impact Search Volume Essential Dummy Gender Dummy Threat
4
METHODOLOGY
This chapter provides an explanation of multi-level regression analysis and how the data was prepared. Furthermore, I will elaborate upon the development of the statistical models for each of the formulated hypotheses and the assumptions that should be satisfied prior to performing the actual analyses.
4.1
Multilevel regression analysis
The dataset that was provided has the structure of panel data (or longitudinal data): participants can evaluate multiple companies per time (cross-sectional dimension) and some participants have participated for multiple years (time series dimension)(Hsiao, 2007). Panel data has the following structure: Xit, i =1,..., N, t =1,…,T, where i is the individual participant and t is the
year. Besides increased accurate inferences of model parameters, a major advantage for using panel data is that it can better model complex human behavior than one would be able to do with cross-section or time-series data (Hsiao, 2007). The dataset is unbalanced because not all participants have evaluated companies for each year in the dataset. Furthermore, the data is gathered continuously, where participants can occur multiple times in the dataset in a non-structured manner. For example, participant a can occur in March 2018, April 2018 and after some time again in January 2020, where participant b only occurs once in June 2020.
In addition, the data follows a cross-classified structure, because each respondent can be nested in multiple companies, and a company can be nested into an industry (Snijders & Bosker, 2012). This is not to be mistaken with a hierarchical structure, where each respondent would only appear once in a higher-level variable such as company. In order to use this type of structure in multilevel modeling, I first specify the random intercept model, where the random variables that represent random differences between respondents, companies and industries are included as intercepts:
𝑌𝑖𝑗 = 𝛽0𝑖𝑗𝑘+ 𝛽1𝑥𝑖𝑗+ 𝑅𝑖𝑗 (1) Where 𝛽0𝑖𝑗𝑘 is the random intercept that includes the average intercept 𝛾00 and the group-dependent deviation 𝑢0𝑖𝑗𝑘:
the number of parameters that are estimated in a model. Therefore, the correct number of degrees of freedom can be determined for the estimation of the random components. The R package “lmer4” with the “lmer” function used to estimate the models with customer satisfaction as dependent variable, and the “glmer” function for the model with recommendation as dependent variable. This resulted in several different model specifications.
4.2
Data preparation
Prior to developing and specifying the final models, the data was scanned for missings and multicollinearity, after which some changes were made that will be further explained in this section.
Out of all the variables, only the variable ‘Income’ shows missings, which is likely the result of people not wanting to share their income with the data provider. In order to deal these missings, the same approach as by de Haan et al., (2015) was used. The missings were set to the ‘average’ value, which is the Dutch modal income. In order to indicate that respondents had not stated their income, a dummy variable was created that takes on a value of 1 if the respondent had not provided his or her income. This allows for a comparison whether the group of respondents that has not provided their income differs significantly from the group that has.
Thereafter, the data was checked for multicollinearity by looking at the Variance Inflation Factors (VIFS). When building models, it is assumed that there is no multicollinearity among the variables in the dataset (VIFs <10), because multicollinearity in the data could result in biased or faulty estimates (Leeflang, Wieringa, Bijmolt, & Pauwels, 2015). In the model estimating recommendations, multicollinearity was not an issue. However, multicollinearity was an issue in estimating customer satisfaction, so therefore a remedy had to be found.
Furthermore, only the crisis variable ‘Hospitalizations’ was kept in the final model for two reasons. First, when all of the crisis variables were included in the model, this resulted in very high VIF statistics (see Appendix 8) for the interactions of CSR perception with ‘StringencyIndex’, ‘SearchVolume’, and ‘ExpectedEconomicImpact’, as well as the CSR perception variable on its own. Because the four crisis variables are all variables that measure the depth of the crisis (and thus move into the same direction, which could explain the multicollinearity), it would make sense to remove the ones that cause multicollinearity. Second, none of the estimates of the crisis variables (except for hospitalizations) and interactions with CSR perception were significant (see Appendix 9). Therefore, for the sake of building a parsimonious model, it was decided to drop the variables that had no explanatory power in order to increase model fit.
Finally, for investigating the three-way interactions between the crisis relevance variables, crisis depth and CSR perception in relation to customer satisfaction, it was decided to estimate the corresponding models separately because including all three of the three-way interaction effects in one model resulted in multicollinearity.
4.3
Model development
Section 4.2 explained the structure of the data and the type of models that will be used for the analyses. How the respondents in the dataset rate the service of the company, will be used as a proxy for customer satisfaction. Because the data follows a cross-classified structure, I specify a multilevel regression model in order to account for the three levels in my data: respondents nested in companies and companies nested in industries. Therefore, for each of the levels, a random intercept is included in the model for respondent, company and industry1. The models
will be estimated for the period of the 29th of February (day of the first hospitalization) until the
5th of October (last date in the Market Response dataset), in order to make an equal comparison
between each of the years. The In order to test whether a crisis impacts customer satisfaction, first a model was estimated on the complete dataset that does not include the crisis variables.
1 In order to test whether it is appropriate to include a random intercept, the Hausman test (1978) was performed
The model includes interactions of year (2019 and 2020) with CSR perception. The following model specification includes all of the regressors (except crisis variables) and interaction effects of ‘Year2019’ and ‘Year2020’ with ‘PerceptionCSR’ on the dependent variable:
𝐶𝑆𝑖𝑗𝑘𝑡= 𝛽0𝑖𝑗𝑘+ 𝛽1𝑃𝑒𝑟𝑐𝑒𝑝𝑡𝑖𝑜𝑛𝐶𝑆𝑅𝑖𝑗𝑘𝑡+ 𝛽2 𝐺𝑒𝑛𝑑𝑒𝑟𝐷𝑢𝑚𝑚𝑦𝑖𝑗𝑘+ 𝛽3 𝐸𝑠𝑠𝑒𝑛𝑡𝑖𝑎𝑙𝐷𝑢𝑚𝑚𝑦𝑗+ 𝛽4𝑇ℎ𝑟𝑒𝑎𝑡𝐷𝑢𝑚𝑚𝑦𝑖𝑗𝑘+ 𝛽5𝑄1 + 𝛽6𝑄3 + 𝛽7 𝑄4 + 𝛽8𝑅𝑒𝑔𝑖𝑜𝑛𝑖+ 𝛽9𝐼𝑛𝑐𝑜𝑚𝑒𝑖𝑗𝑘+ 𝛽10𝐼𝑛𝑐𝑜𝑚𝑒𝑁𝑎𝐷𝑢𝑚𝑚𝑦𝑖𝑗𝑘+ 𝛽12𝑌𝑒𝑎𝑟2019 + 𝛽13𝑌𝑒𝑎𝑟2020 + 𝛽14(𝑌𝑒𝑎𝑟2019 ∗ 𝑃𝑒𝑟𝑐𝑒𝑝𝑡𝑖𝑜𝑛𝐶𝑆𝑅𝑖𝑗𝑘𝑡) + 𝛽15(𝑌𝑒𝑎𝑟2020 ∗ 𝑃𝑒𝑟𝑐𝑒𝑝𝑡𝑖𝑜𝑛𝐶𝑆𝑅𝑖𝑗𝑘𝑡) + 𝑒𝑖𝑗𝑘 (3) with 𝛽0𝑖𝑗𝑘= 𝛾00𝑘+ 𝑢0𝑖𝑗𝑘 and 𝛾00= 𝛽000+ 𝑢00𝑘 Where;
CSijkt customer satisfaction for respondent i, firm t, industry j at time t
PerceptionCSRijkt individual CSR perception of respondent i for firm t in industry j at time t
(1= respondent perceives firm to engage in CSR) GenderDummy gender of respondent i (1= male)
EssentialDummy industry type of company j (1= Essential firm) ThreatDummy dummy variable (1= respondent is 70 years or older) Q1 dummy variable (1= observation is in the 1st quarter)
Q3 dummy variable (1= observation is in the3rd quarter)
Q4 dummy variable (1= observation is in the 4th quarter)
Region region where someone is from (worst hit, medium hit, least hit) Income income of customer i (above average, average, below average) IncomeNaDummy dummy variable (1= income was not known for respondent i)
Year 2019 dummy variable (1= year 2019)
Year 2020 dummy variable (1= year 2020)
𝑒𝑖𝑗𝑘 error term
If the interaction effects are significant (p<0.05), I can assume that there is a significant difference in the effectiveness of CSR perception on customer satisfaction between 2018, 2019 and 2020. Here, 15 is thus used to test hypothesis H3a, which is expected to be positive.
Furthermore, the effect of CSR perception (1) on customer satisfaction is used to test H1, of
