A multilevel regression approach to understand the impact of the
Corona-virus crisis on the service level and brand evaluations of Dutch supermarkets.
Master's Thesis Marketing Intelligence
EBM867B20
Flip Ruys
S2763702
Abstract
The consequences of the lock down and social distancing policies initiated by governments have made noticeable changes on our daily lives. Due to the Corona crisis, certain drivers of brand evaluations or customer satisfaction seem to have changed. This research discusses the effect of COVID-19 on cus-tomer satisfaction. More specifically, cuscus-tomer satisfaction was measured by perceived service qual-ity. Survey respondents expressed their satisfaction by evaluating Dutch supermarkets. In marketing, there is a substantial lack of research about the effects of pandemics on customer evaluations. This re-search introduces a multilevel regression model and aims to contribute insights into the consequences of this pandemic as regards different service quality levels measurable via customer service quality experiences and evaluations. Overall, the results show that during the first wave of COVID-19 infec-tions, customer satisfaction declined. However, considering 2020 as whole, the effect of the corona-virus crisis differed depending on the perceived service quality dimension.
1. Introduction
During the past decades, many conceptual and empirical studies have examined crises, recessions and depressions. Previous studies have shown that retailers suffer due to less consumption during economic downturns (Antonio et al., 2019; Arguello et al., 2020; Farber, 2015). In a financial and economic crisis, households face high uncertainty about their future income and the uncertain economic environment, which leads to a more conservative approach concerning expenditures (Antonio et al., 2019). The most recent crisis, the financial crisis of 2008, had a negative effect on multiple sectors (Arguello et al., 2020). Specifically, in the United States, the rate of job losses increased by almost 2% in the years after the crisis (2007-2009) compared to 2005 (Farber, 2015). Due to recent comparable crisis conditions, spe-cifically the effects of COVID-19, millions of people are in danger of unemployment. The International Labor Organization has reported that globally, approximately 25 million jobs are expected to be lost due to the coronavirus crisis (ILO, 2020).
Since the end of 2019, the COVID-19 pandemic has caused serious health and financial consequences all over the world. This infectious disease is caused by severe acute respiratory syndrome (SARS-CoV-2) and can result in severe symptoms (Phelan et al., 2020). The disease has been reported in 188 coun-tries and caused close to 82.7 million cases of infection, with that number growing every day (Worldometer, 2020). In addition, the negative financial impact is also growing every day. In 2021 global growth is projected at 5.4 percent. Overall, this would leave 2021 GDP some 6½ percentage points lower than in the pre-COVID-19 projections of January 2020 (IMF, 2020). According to the International Monetary Fund (IMF, 2020), the decline in the gross domestic product (GDP) shows that the coronavirus crisis is the worst crisis since the Great Depression of the 1930s. Figure 1 shows an 8.5% decline in the GDP of the Netherlands.
As a consequence, many stores have filed for bankruptcy, for instance, in the catering industry (CBS, 2020).
Figure 1 Gross domestic product of the Netherlands (CBS, 2020).
In these turbulent times, when customers appear to be more committed to social distancing and sanitised retail entrances (Rukuni & Maziriri, 2020), individuals – especially those at risk of becoming easily infected – want to avoid this contagious disease while still feeling comfortable when shopping. Con-sumers’ expect retailers to adapt to new regulations and offer extra services. Therefore, keeping a good
relationship with customers is essential. Listening to them and providing the expected high service level could be of utter importance to overcome crises. As a result of the crisis, companies are adapting their goals and helping to tackle the pandemic; for example, local breweries have started to produce hand sanitiser. These activities are called “purpose marketing” (Hoekstra & Leeflang, 2020). The importance of service and the significance of perceived service quality have become increasingly relevant (Dehghanpouri et al., 2020). Increased competition, highly informed customers and higher standards of living have led companies to revise their customer service strategies (Arora & Narula, 2018). To battle the growing homogenisation of available products in the market and to create a differentiated advantage in the face of competition, companies must ensure service quality (Kaura et al., 2015). According to Kumar and Reinartz (2016), the essence of great service is that it generates customer satisfaction, which leads to a higher customer lifetime value (CLV). The CLV indicates how valuable a customer is; more specifically, it explains the total revenue generated over the life of the customer for the firm. In the United States, the CLV for a family is estimated to be US$ 500,000 over 25 years for supermarkets (Sheth et al., 2020). To enhance CLV figures, companies adopt new strategies that improve their rela-tionships with customers. The existing literature (Arguello et al., 2020; Kaura et al., 2015; Parasuraman et al., 1985) has described the variables associated with high perceived service quality.
market entrants (Herbane, 2013). In the retail industry, these opportunities are exploited by new internet-based retailers due to recent developments in e-commerce. New companies enter the retail grocery mar-ket by providing online supermarmar-kets that fulfil home deliveries from their warehouses (Pan et al., 2017). Not only supermarket chains have been moving quickly to add e-commerce, but firms such as Amazon acquire supermarket chains (Whole Foods) to explore grocery delivery services (Lu et al., 2013). As a result of a crisis or price war, a growing number of physical stores have to close, especially smaller market players; these closures result in opportunities for non-physical competitors such as Amazon or Bol.com. The latter, a Dutch online retailer, reported significant revenue growth of nearly 30%, or €2.8 billion, in 2019 (Ahold Delhaize, 2020).
In marketing, various studies have demonstrated that brand evaluations are one of the best ways to ana-lyse a brand (Chang & Chieng, 2006; Dyson, 1996; Keller, 2009). However, most such studies have only investigated the effect on performance or firm value (Luo et al., 2013). Additionally, the relation-ship between consumers' evaluations and possible brand extensions has been widely addressed (Jin & Zou, 2013; Dall’Olmo Riley et al., 2014). However, a gap exists in relation to customer evaluations. Marketing managers require guidance concerning the perceptions of their customers. To retain custom-ers and keep them satisfied, marketing managcustom-ers must pay more attention to their customcustom-ers' opinions (Otto et al, 2020) – especially regarding COVID-19 and how customers think the firm should react. In marketing, there is a substantial lack of research about the effects of a pandemic on customer evalua-tions. This research aims to provide insight into the consequences of the pandemic on different customer satisfaction, measurable by perceived quality dimensions of Dutch supermarkets.
2. Literature Review
2.1 Crises
Every crisis has the same characteristics. First, there is a threat with catastrophic effects. Second is the element of surprise. Third, there is a short decision time for governments or the concerned management team, that experience such crises, to attempt implementations. Last, the incurred damage must be recov-ered (Hagen et al., 2013). Although crises have the same structure, they noticeably differ. For example, a natural disaster represents a “hazard cycle in which agencies try to decrease vulnerability by taking measures to reduce casualties and exposure to damage and disruption” (Hagen et al., 2013). A financial crisis is “a disruption of financial markets in which adverse selection and moral hazard problems become much worse, making financial markets unable to efficiently channel funds to those who have the most productive investment opportunities” (Mishkin, 1992). To better understand the impact of different types of crises, the following sections discuss examples of such crises: the Great Recession and the coronavirus crisis.
2.1.1 The Great Recession
The global economic and financial crisis that began at the end of 2007 had a tragic effect on the econ-omy, housing industry, retail, firms and households. Prior to the recession, there was a period of eco-nomic growth and stability in many countries. According to Grusky et al. (2011), the most influential cause was the beginning of the mortgage securitisation industry. What emerged in a relatively small niche market quickly started to grow and eventually was generating 40% of the profits in the American economy as a whole. These profits were particularly generated from businesses engaged in selling mort-gages and creating different forms of mortgage-backed securities and related financial products. In 2007, the housing and mortgage industry collapsed, resulting in other financial disasters. The mort-gage industry had become a structure that linked all of the financial institutions and made them depend-ent on house prices. Among the recession’s roots were the availability of easy credit, overexpansion of companies, outsourcing and sharp increases in energy prices (Hetzel, 2011). In reaction to the uncer-tainty, consumers and firms substantially reduced their purchasing.
Figure 2 visualises the significant 4.5% decline in the Dutch gross domestic product during the Great recession and the post-cri-sis years.
Figure 2 Gross domestic product of the Netherlands (CBS, 2016). While the Great Recession unquestionably changed the economic environment, it also fundamentally changed the behaviour of consumers. In addition to the decline in expenditures, consumers seemed to demand simplicity. The recession also created a desire for trusted brands, excellent service and advisers able to simplify decision-making (Flatters & Willmott, 2009). The Great Recession illustrated the im-portance of listening to consumers to prevent bankruptcy and provide shoppers with excellent service.
2.1.2 The Coronavirus crisis
Most research on the Coronavirus crisis has sought to conceptualise the holistic impact of the pandemic, with the human toll of the virus expressed in financial and psychological terms (Crayne, 2020; Liu et al., 2020). The current and future pandemic impact on the global economy has been widespread, includ-ing a potential recession that would put both entire economies and numerous industries at risk (Liu et al., 2020). Apart from the financial consequences, the pandemic has also had physical and psychological effects.
Accordingly, the World Health Organization and local governments all over the world have encouraged social distancing, hand washing and mask wearing to avoid infection (Goh et al., 2020). Although re-search suggests that these measures effectively reduce the spread of the virus, they also appear to have psychological drawbacks. Mental health professionals have expressed their concern that quarantine leads to a greater risk of depression; continuously hearing and reading about COVID-19 can also be stressful (Di Renzo et al., 2020). Social distancing and isolation are correlated with psychological dis-tress and have a negative effect on psychological well-being (Best et al., 2020).
between brands’ images and actual brand evaluations in the automotive industry. The survey points out that customers start to value tangible goods in economic downturns. Due to uncertainty during and after the recession in the automotive industry, customers had higher expectations. Rather than just focusing on the driving experience, customers also showed interest in service quality concerning the purchase itself (experience of shopping), the dealership and ethical credentials (Handly, 2010). As a result of higher expectations from customers, brand evaluations will be less positive during downturns if firms do not adapt to these expectations.
In reaction to current crisis, consumers’ economic certainty has decreased, which has consequences across industries. One of the largest industries that experiences these consequences and sell goods or services directly to consumers or end-users, is the retail industry. Supermarkets can be divided into three segments: service, middle and discount (Distrifood.nl, 2020).
Service: This category consists of supermarkets that provide customers with high service levels and wide assortments at a relatively high price per product.
Middle: Supermarkets in this category feature medium-level prices and a medium service level. Discount: Supermarkets in this category offer products at relatively low prices and with a low service level.
Figure 3 illustrates the three strategies discussed above. The X-axis indicates the perceived service level, which spans from very poor (left) to excellent (right). The Y-axis describes the discount level. Grocery stores towards the bottom of the graph offer slight discounts. In the matrix, the relative positions of all major Dutch grocery stores are presented.
2.2 Brand evaluations
As customers become more educated, managers face the challenging task of coping with continuously changing customer demands and maintaining a persistent brand image. In the past, retailers enhanced service quality in an effort to boost customer perceptions and evaluations (White, 2013). For consum-ers in the traditional retail industry, evaluations of service quality rely on the following subdimen-sions: the interaction, the physical environment and the outcome (Brady, 2001). The goal for market-ing managers is to continuously improve these subdimensions to create an excellent service experi-ence. To understand the psychological state of customers and their experiences, companies implement brand evaluations to obtain more insight into customer satisfaction with their brands. These evalua-tions demonstrate the brand experience and how the brand is interpreted by customers. Brand
experi-ences are the “subjective, internal consumer responses (sensations, feelings, and cognition) and
behav-ioral responses evoked by brand-related stimuli that are part of a brand’s design and identity, packag-ing, communications and environments” (Brakus et al., 2009). A brand evaluation offers a better un-derstanding of overall customer satisfaction with the brand and can possibly lead to indicators of fu-ture business performance.
Four factors affect customer buying decisions (Khan et al., 2014): external stimuli, internal perceptions, buying behaviour and demographic variables. Internal perceptions describe the lifestyle, emotions and feelings of the customer. In the retail industry, marketeers continuously try to anticipate these four di-mensions to realise a higher customer value. Moreover, positive supermarket experiences stimulate con-sumers’ senses and engage them through emotion, cognition and similar positive experiences (Dolbec & Chebat, 2013).
In service evaluations across brands, customers assess service performance in terms of expectations and acceptance. What customers expect and what they deem acceptable separate their desired outcome from the perceived service level (Dlamini & Bernard, 2020). For retailers, customer expectations are imper-ative to understanding the desired perceived quality of shoppers. Moreover, this variable is how the customer experience and customer satisfaction are measured (Carpenter, 2006).
highly beneficial for the involved brands; demand was so strong that suppliers barely could keep up. Temporary price reductions seem highly effective; however, some brands explicitly choose to stay out of price wars (Van Aalst et al., 2005). To make a temporary price reduction effective, high revenue needs to be realised to compensate for the smaller profit margin. For example, Campina, the market leader for most Dutch dairy products, increased prices to stay out of aggressive price wars (Van Aalst et al., 2005). Starting a price war can cause a change in how customers interpret the brand. Consumers assess certain brands in their minds, and they evaluate them using the general attributes of these brands (Veloutsou, 2015). The ultimate goal of marketers is to positively influence consumers regarding their brands. Satisfied customers are namely widely recognised as a key indicator of a company’s success since they imply positive brand attitudes and purchases (Spreng et al., 1995; Oliver, 1980; Grace & O’Cass, 2005).
However, while drivers of customer satisfaction have been researched in multiple contexts, few scholars have investigated the effect of a crisis, let alone a pandemic, on customer satisfaction in the retail indus-try. Nevertheless, some drivers of brand evaluations or customer satisfaction seem to have changed during the coronavirus crisis. According to a recent experiment that examined the effects of norm vio-lations (e.g., social distancing) in a grocery store setting, customer satisfaction was negatively affected by employees that violated Corona measures (Söderlund, 2020). The participants were exposed to a grocery store employee whose behaviour was manipulated. Not surprisingly, norm-violating behaviour generated a higher level of disgust towards the employee in question and led an overall lower score on the evaluation of the supermarket.
Figure 4 Conceptual model.
2.3 Service level
Over time, various researchers and companies have expressed interest in the level of service quality. Due to growing competition, many firms in the retail industry seek profitable ways to differentiate them-selves. Superior service quality is part of a firm’s strategy to succeed and to create a greater market share (Solimun, & Fernandes, 2018). Unlike product quality, which can be measured objectively, service quality is abstract and intangible. For this reason, scholars find it difficult to define and measure service quality (Gavin, 1983; Grönroos, 1984; Parasuraman, 1988).
The initial SERVQUAL scale was introduced as a 10-dimensional measurement of service quality, but later reduced to five dimensions: reliability, assurance, tangibility, responsiveness and empathy (Par-asuraman et al., 1985, 1988). The authors argued that the gap between what customers expect from a service and their perception of the actual service is essential to capture in a valid service quality meas-urement instrument (Collins, 2017).
2.4 Hypotheses
For this research, customer service was measured by six dimensions indicating possible changes in per-ceived service quality. Before conducting the research, several hypotheses were formulated. Every hy-pothesis consists of two parts. The first part refers to the general relationship between the service di-mension in question and customer satisfaction. The second part refers to the effect of the coronavirus crisis on that relationship between a service dimension and customer satisfaction.
First, many countries implemented a quarantine that resulted in a temporary or permanent loss of em-ployment (Mimoun et al., 2020). Due to illness of employees as a result of COVID-19, the expectation was that supermarkets struggle with deploying employees to help customers. Therefore, the following hypothesis was formulated:
H1: The positive impact of the availability of supermarket employees on customer satisfaction (1a) has been negatively affected by the coronavirus crisis (1b).
Second, to help customers feel comfortable, supermarkets try to manage the traffic and food supply to improve the shopper experience in their stores. However, in the early weeks of the pandemic, retailers experienced overwhelming demand paired with panic buying, which resulted in empty shelves (Leone et al., 2020). If grocery stores make empty promises in advertisements and customers have unpleasant experiences while shopping, these factors could have a negative effect on the credibility of the brands. H2: The positive impact of the credibility of supermarkets on customer satisfaction (2a) has been neg-atively affected by the coronavirus crisis (2b).
Third, customer–employee contacts are crucial encounters that have a considerable impact on shoppers’ impressions of the retailer. Customers’ interactions with employees could be experiences that lead to favourable consumer reviews (Lucia-Palacior et al., 2020). On the other hand, such recommendations could also irritation in a disturbing way. Due to coronavirus restrictions and social distancing between customers and employers, the expectation was that there has been less interaction and advice from em-ployees.
H3: The positive impact of customers not being unnecessarily disturbed while shopping by supermarket employees on customer satisfaction (3a) has been strengthened by the coronavirus crisis (3b).
Fourth, service quality can be measured in the way that grocery stores handle formalities. In this context, formalities refer to possibility to return goods to the store and receive a refund. There should be a clear procedure that handles these issues. To encourage positive shopping experiences and brand perceptions, firms should communicate in a clear manner. Therefore, one service dimension is that customers expect supermarkets not to complicate formalities.
H4: The positive impact of the way that customer’s experience handled formalities by supermarkets on customer satisfaction (4a) has been positively affected by the coronavirus crisis (4b).
Fifth, an important attribute of perceived service quality is that customers assume that supermarkets admit to and solve problems and mistakes. Grocery stores ought to be transparent and curious about customers’ opinions. A related hypothesis sought to determine if the coronavirus crisis has resulted in a higher number of admitted mistakes and the level of resolved problems.
H5: The positive impact of supermarkets’ problem solving on customer satisfaction (5a) has been pos-itively affected by the coronavirus crisis (5b).
Finally, it was expected that customers would perceive a higher level of involvement due to new coro-navirus measures. Customer involvement describes the level of interaction between the customer and retailer and should be based on in-depth understanding of customers and their specific needs (Kim et al., 2018). Due to new measures, grocery stores have adopted new rules and are more solution-oriented. By listening to their customers, they can achieve a higher level of involvement.
H6: The positive impact of supermarkets’ involvement with their customers on customer satisfaction (6a) has been positively affected by the coronavirus crisis (6b).
This study aims to provide, for the Dutch retail industry, insight into the impact of the coronavirus crisis on different service levels. Specifically, it considers the consequences of the pandemic on brand evalu-ations.
3. Data Description
3.1 Customer satisfaction
cross-sectional- or time series data (Baltagi, 2008). Moreover, panel data models provide ways of deal-ing with heterogeneity and examindeal-ing fixed and/or random effects in longitudinal data (Baltagi, 2008). For the analyses, some demographic variables needed recoding to make interpretation easier. First, the income variable was recoded. This variable contained three groups: “below modal, modal & above modal”. However, due to the relatively high number of “n/a” responses (14.7% of total income), a dummy variable was created. The reason for the creation of the dummy variable is that simply removing the variable would have led to a loss of valuable information. The dummy was set to 1 if the respondent did not provide his or her income (De Haan et al., 2015). Moreover, the “n/a” responses were imputed as the average Dutch modal income. Furthermore, dummies were created to distinguish between the supermarket chains. As discussed before, supermarkets can be divided into three types: service, middle and discount. Additionally, to distinguish peaks and valleys of the impact of the coronavirus crisis, the dataset was divided into three years: 2018, 2019 and 2020.
3.2 Indices of the severity of the Coronavirus crisis
The first COVID-19-related case in the Netherlands was identified on 27 February 2020, and the first COVID-19-related death was on 6 March 2020. In reaction to the growing number of infections, the Dutch government decided to introduce an “intelligent lockdown”. In extension of this new measure, gatherings with over 100 people were banned. Moreover, universities decided to switch to online teach-ing, and most companies supported working from home. Schools and day-care centres were also forced to close their doors temporarily. The Dutch government’s strategy of a nationwide lockdown, which was initiated 12 March 2020, had an unexpected outcome. The incidence of the illness and its timing were found to vary substantially across the provinces (RIVM, 2020).
Figure 5 provides the number of hospital admissions per day in the Netherlands. The figure makes a comparison between the provinces that were the least and the most affected by the virus.
corona-3.2.1 Hospital admissions
Every day, hospitals register data from COVID-19 patients from the nursing and intensive care departments. Figure 6 shows the number of hospital admissions per day due to coronavirus infections (RIVM, 2020).
Figure 6 Dutch hospital admissions in 2020 (RIVM, 2020). The Y-axis displays the number of admissions and the X-axis the exact date in 2020. Data was available from 27 February until 5 October and reflects patients in the Netherlands.
3.2.2 Number of casualties
The Dutch Area Health Authority also reg-isters the number of casualties due the COVID-19 virus. Figure 7 shows the num-ber of deaths per day due to coronavirus in-fections (RIVM, 2020).
Figure 7 Number of coronavirus casualties in the Netherlands in 2020 (RIVM, 2020). The Y-axis displays the number of casualties and the X-axis the exact date in 2020. This data was avail-able from 27 February until 5 October.
3.2.3 Stringency index
Across the globe, governments have taken different measures in response to the pandemic. To track and compare policy responses, the University of Oxford has created a tool. The tool is called the Oxford COVID-19 Government Response Tracker (OxCGRT) and systemically collects information on differ-ent policy responses that governmdiffer-ents have taken in response to the COVID-19 outbreak (Hale et al., 2020). The tool consists of 18 indicators such as school closures and travel restrictions.
These indicators are divided into four common subgroups, and scores vary between 1 and 100 to reflect the level of government action. The four subgroups are described below:
2. Containment and health index: combines “lockdown” restrictions and closures with measures such as testing policies and contact tracing, short-term investments in healthcare and invest-ments in vaccines.
3. Economic support index: measures such as income support and debt relief. 4. Original stringency index: the strictness of “lockdown-type” policies
that primarily regulate people’s behaviour.
All four dimensions were merged into one general index, the stringency index (OcCGRT), and gave more insight into the impact of the coronavirus crisis on the perceived service quality in Dutch super-markets during the pandemic.
3.2.4 Online search behaviour
After the outbreak of the coronavirus, much online research considered the spread of the
vi-rus. Google Trends allows one to view the relative search frequencies of particular search
terms (Szmuda et al., 2020). By examining search terms linked to the virus, one can use
Google Trends as a tool to measure the impact of the pandemic. The search terms included
were “corona”, “corona virus” and “corona viruses”.
3.3 Measurements of the service dimension
It is common to use dummy variables in marketing research, for example, in regression analyses. A dummy variable is a numerical variable and is often used to distinguish treatment groups (Worch et al., 2010). The dataset used for this research consisted of multiple dummy variables: the service quality dimensions that indicate the level of agreement with certain statements. The included golden rule (GR) questions considered whether companies were performing well in the eyes of customers. An example is the following statement: “There are plenty employees available that are willing to help”. The respondent had only two answer options: “do not agree” (0) and “agree” (1). According to the respondent, a com-pany had complied (or not) with the service quality standards of the customer.
3.4 Measurements of evaluation
Brand evaluations were measured through a customer satisfaction score. As stated before, customer expectations are imperative to understanding the desired perceived quality of shoppers. Moreover, cus-tomer satisfaction represents cuscus-tomers’ evaluations and expectations (Carpenter, 2006). In this re-search, a customer satisfaction score from 1 to 10 was included. A score of 1 represented a customer who was not satisfied, while 10 represented a highly satisfied customer.
Table 1 Variable explanation.
The most traditional approach to check for collinearity is to examine a correlation matrix. Highly posi-tive or negaposi-tive correlations are indications of potential difficulties in the estimation of reliable effects (Leeflang et al., 2015). The correlation matrix in Table 2 provides initial insights. Multicollinearity occurs when two or more variables on the right-hand side of a regression model are highly correlated (Disatnik & Sivan, 2016). The correlation matrix (Table 2) shows that there was no multicollinearity due relatively low scores. Also, Table 3 shows descriptive statistics.
Table 2 Correlation Matrix.
Variable Definition
Respondent ID A unique value for every respondent Province Province in the Netherlands
Gender The sex of the respondent (male, female)
Age The age of the respondent
Income The income of the respondent (below modal, modal, above modal) Company Supermarket chain in the Netherlands
GR1: Employees GR1: Availability of employees (1 = agree, 0 = not agree) GR2: Credibility GR2: Credibility for customers (1 = agree, 0 = not agree) GR3: Contact GR3: No unnecessary contact (1 = agree, 0 = not agree) GR4: Formalities GR4: Handled formalities (1 = agree, 0 = not agree)
GR5: Mistakes GR5: Admitting mistakes and solving problems (1 = agree, 0 = not agree)
GR6: Involvement GR6: Level of involvement (1 = agree, 0 = not agree)
Satisfaction Overall satisfaction score (1 = very unsatisfied, 10 = very satisfied) Recommendation Recommendation of supermarket (1 = yes, 0 = no)
Stringency index Index that represents common policy responses that governments have taken to respond to the pandemic
Hospital admission Number of hospital admissions per day Deaths last week Number of deaths of past week per day
Search term index Index of search interest in “corona”, “corona virus” and “corona-viruses” (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (1) GR1: Employees 1 (2) GR2: Credibility 0.1 1 (3) GR3: Contact 0.1 0.4 1 (4) GR4: Formalities 0.2 0.5 0.4 1 (5) GR5: Mistakes 0.2 0.4 0.4 0.3 1 (6) GR6: Involvement 0.2 0.5 0.3 0.4 0.4 1 (7) Satisfaction 0.2 0.3 0.2 0.2 0.3 0.4 1 (8) Recommendation 0.1 0.2 0.3 0.3 0.2 0.4 0.2 1 (9) Stringency Index 0.0 0.0 -0.3 -0.2 -0.1 0.0 0.0 -0.1 1 (10) Hospital Admis. 0.0 0.0 -0.1 0.0 0.0 0.0 0.0 0.0 0.2 1
(11) Deaths last week 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.4 1
Table 3 Descriptive statistics (n = 10,934). M (SD) or % Range Categorical Variables Gender Male 42.2% Female 57.8.% Income Below modal 49.1% Modal 18.5% Above modal 17.5% Declined to answer 14.9% Province Zuid-Holland 19.8% Noord-Brabant 16.8% Noord-Holland 16.0% Gelderland 10.6% Limburg 8.2% Utrecht 7.1% (Other) 21.5% Company Albert Heijn 24.1% Jumbo 16.5% Lidl 16.3% Aldi 10.0% Plus 6.4% Coop 4.7% (Other) 22%
Golden Rule 1: Employees 0.575 0–1
Golden Rule 2: Credibility 0.635 0–1
Golden Rule 3: Contact 0.413 0–1
Golden Rule 4: Formalities 0.645 0–1
Golden Rule 5: Mistakes 0.534 0–1
Golden Rule 6: Involvement 0.412 0–1
Recommendation 0.360 0–1
Numeric variables
Satisfaction 7.645 (1.65) 1–10
Stringency index 46.74 (35.37) 1–79.63
Hospital admission 0.046 (0.60) 0–30
Deaths last week 9.072 (0.46) 0–175
4. Methodology
4.1 Multilevel regression
Given that this study’s research objective was to determine the consequences of a crisis on customer satisfaction and evaluations and uses panel data, a multilevel regression model was estimated. Tradi-tional analyses of satisfaction neglecting the multilevel structure of a dataset usually lead to questionable estimations and inferences (Tso & Guan, 2014). The drivers of such questionable estimations could be the inconsistency between the multilevel structure and assumptions, such as the fact that observations are independently and similarly distributed. (Tso & Guan, 2014). Traditional methods, such as multiple linear regression analysis, are based on observations that are not associated with each other. However, there is a non-ignorable change that this assumption does not hold. This implies that observations in the supermarket evaluation survey are related due to the multilevel structure. For this reason, traditional methods are not applicable in all situations. According to Gelman et al. (2007), an advantage of a mul-tilevel model is that it copes with the heterogeneity of relationships between explanatory variables and response variables within the data. Due to the multilevel model, the heterogeneity of relationships that induce debatable outcomes was ruled out.
4.2 Model estimation
Since the panel data had a hierarchical multilevel structure, a multilevel regression analysis was per-formed. The multilevel regression was conducted with random intercepts at the respondent ID and company level. To explain differences between the years before and during the pandemic, and the ef-fects of variables on customer satisfaction, two models were built. The first model, specified by Equa-tion 1, compares the calculated customer satisfacEqua-tion for 2018, 2019 and 2020. The year 2018 was se-lected as the baseline and compared to 2019 and 2020 to identify potential effects. Moreover, if certain variables showed significance in 2020 in comparison with 2018, the coronavirus could be an indicator for a different customer satisfaction score. To calculate the effects of the variables: stringency index,
hospital admission, deaths last week and search term during the pandemic, a second model was built.
Model 2, specified by Equation 2, explains how the included variables affected customer satisfaction in 2020. Table 4 explains the variables in these two equations. For example, the parameter 𝛽! is the effect on customer satisfaction when a customer is one year older, on the corresponding Age variable
Table 4 Variable definitions.
Variable Description
𝐴𝑔𝑒",$ The age of the respondent i on day t
𝐹𝑒𝑚𝑎𝑙𝑒",$ Dummy indicating whether respondent i on day t was female
𝐼𝑛𝑐𝑜𝑚𝑒",$ The income of respondent i on day t
𝐼𝑛𝑐𝑜𝑚𝑒. 𝑛𝑎",$ Dummy indicating whether the income of respondent i on day t is un-known
𝑃𝑟𝑜𝑣𝑖𝑛𝑐𝑒",$ Province where the residence of respondent i was located on day t
𝑌𝑒𝑎𝑟$ Year 2018, 2019 or 2020 on day t
𝑆𝑒𝑔𝑚𝑒𝑛𝑡. 𝑑𝑖𝑠𝑐𝑜𝑢𝑛𝑡",$ Dummy indicating discount supermarkets (Aldi, Boni, Dirk van den Broek, Lidl,Nettorama and Vomar) that respondent i visited on day t 𝑆𝑒𝑔𝑚𝑒𝑛𝑡. 𝑠𝑒𝑟𝑣𝑖𝑐𝑒",$ Dummy indicating service supermarkets (Albert Heijn, Jumbo, Picnic,
Plus, Poiesz and Spar) that respondent i visited on day t
𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑒𝑠",%,$ How respondent i perceived the service dimension “availability of em-ployees” for company j on day t
𝐶𝑟𝑒𝑑𝑖𝑏𝑖𝑡𝑙𝑖𝑡𝑦",%,$ How respondent i perceived the service dimension “credibility for cus-tomers” for company j on day t
𝐶𝑜𝑛𝑡𝑎𝑐𝑡",%,$ How respondent i perceived the service dimension “no unnecessary contact” for company j on day t
𝐹𝑜𝑟𝑚𝑎𝑙𝑖𝑡𝑖𝑒𝑠",%,$ How respondent i perceived the service dimension “handled formali-ties” for company j on day t
𝑀𝑖𝑠𝑡𝑎𝑘𝑒𝑠",%,$ How respondent i perceived the service dimension “admitting mistakes and solving problems” for company j on day t
𝐼𝑛𝑣𝑜𝑙𝑣𝑒𝑚𝑒𝑛𝑡",%,$ How respondent i perceived the service dimension “level of
involve-ment” for company j on day t
𝑆𝑎𝑡𝑖𝑠𝑓𝑎𝑐𝑡𝑖𝑜𝑛",%,$ Overall satisfaction score for respondent i for company j on day t
𝑆𝑎𝑡𝑖𝑠𝑓𝑎𝑐𝑡𝑖𝑜𝑛2020",%,$ Overall satisfaction score for respondent i for company j on day t in
2020
𝑅𝑒𝑐𝑜𝑚𝑚𝑒𝑛𝑑𝑎𝑡𝑖𝑜𝑛",%,$ “How likely is it that you [respondent i] would recommend [company j
] to a friend or colleague?” (0 = very unlikely, 10 = very likely) on day
t
𝑆𝑡𝑟𝑖𝑛𝑔𝑒𝑛𝑐𝑦_𝑖𝑛𝑑𝑒𝑥$ Index that represents common policy responses that governments have
taken to respond to the pandemic for day t 𝐻𝑜𝑠𝑝𝑖𝑡𝑎𝑙_𝑎𝑑𝑚𝑖𝑠𝑠𝑖𝑜𝑛$ Number of hospital admissions on day t 𝐷𝑒𝑎𝑡ℎ𝑠_𝑙𝑎𝑠𝑡_𝑤𝑒𝑒𝑘$ Number of deaths in past week on day t
Equation 1 Satisfaction in 2018, 2019 and 2020. 𝑆𝑎𝑡𝑖𝑠𝑓𝑎𝑐𝑡𝑖𝑜𝑛",%,$ = 𝛽'+ 𝛽!𝐴𝑔𝑒",$+ 𝛽( 𝐹𝑒𝑚𝑎𝑙𝑒",$+ 𝛽) 𝐼𝑛𝑐𝑜𝑚𝑒",$+ 𝛽* 𝐼𝑛𝑐𝑜𝑚𝑒. 𝑛𝑎",$ + 𝛽+ 𝑆𝑒𝑔𝑚𝑒𝑛𝑡. 𝑑𝑖𝑠𝑐𝑜𝑢𝑛𝑡",$+ 𝛽, 𝑆𝑒𝑔𝑚𝑒𝑛𝑡. 𝑠𝑒𝑟𝑣𝑖𝑐𝑒",$+ 𝛽- 𝑃𝑟𝑜𝑣𝑖𝑛𝑐𝑒",$ + 𝛽. 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑒𝑠",%,$+ 𝛽/ 𝑌𝑒𝑎𝑟$+ 𝛽!' 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑒𝑠",%,$𝑌𝑒𝑎𝑟$+ 𝛽!! 𝐶𝑟𝑒𝑑𝑖𝑏𝑖𝑡𝑙𝑖𝑡𝑦",%,$ + 𝛽!( 𝐶𝑟𝑒𝑑𝑖𝑏𝑖𝑙𝑖𝑡𝑦",%,$𝑌𝑒𝑎𝑟$+ 𝛽!) 𝐶𝑜𝑛𝑡𝑎𝑐𝑡",%,$+ 𝛽!* 𝐶𝑜𝑛𝑡𝑎𝑐𝑡",%,$𝑌𝑒𝑎𝑟$ + 𝛽!+ 𝐹𝑜𝑟𝑚𝑎𝑙𝑖𝑡𝑖𝑒𝑠",%,$+ 𝛽!, 𝐹𝑜𝑟𝑚𝑎𝑙𝑖𝑡𝑖𝑒𝑠",%,$𝑌𝑒𝑎𝑟$+ 𝛽!- 𝑀𝑖𝑠𝑡𝑎𝑘𝑒𝑠",%,$ + 𝛽!. 𝑀𝑖𝑠𝑡𝑎𝑘𝑒𝑠",%,$𝑌𝑒𝑎𝑟$+ 𝛽!/ 𝐼𝑛𝑣𝑜𝑙𝑣𝑒𝑚𝑒𝑛𝑡",%,$+ 𝛽(' 𝐼𝑛𝑣𝑜𝑙𝑣𝑒𝑚𝑒𝑛𝑡",%,$𝑌𝑒𝑎𝑟$ + 𝛽(! 𝑅𝑒𝑐𝑜𝑚𝑚𝑒𝑛𝑑𝑎𝑡𝑖𝑜𝑛",%,$+ 𝛽(( 𝑅𝑒𝑐𝑜𝑚𝑚𝑒𝑛𝑑𝑎𝑡𝑖𝑜𝑛",%,$𝑌𝑒𝑎𝑟$+ 𝜖",%,$ Equation 2 Satisfaction in 2020 𝑆𝑎𝑡𝑖𝑠𝑓𝑎𝑐𝑡𝑖𝑜𝑛2020",%,$ = 𝛽'+ 𝛽!𝐴𝑔𝑒",$+ 𝛽( 𝐹𝑒𝑚𝑎𝑙𝑒",$+ 𝛽) 𝐼𝑛𝑐𝑜𝑚𝑒",$+ 𝛽* 𝐼𝑛𝑐𝑜𝑚𝑒. 𝑛𝑎",$ + 𝛽+ 𝑆𝑒𝑔𝑚𝑒𝑛𝑡. 𝑑𝑖𝑠𝑐𝑜𝑢𝑛𝑡",$+ 𝛽, 𝑆𝑒𝑔𝑚𝑒𝑛𝑡. 𝑠𝑒𝑟𝑣𝑖𝑐𝑒",$+ 𝛽- 𝑃𝑟𝑜𝑣𝑖𝑛𝑐𝑒",$ + 𝛽. 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑒𝑠",%,$+ 𝛽/ 𝑆𝑡𝑟𝑖𝑛𝑔𝑒𝑛𝑐𝑦_𝑖𝑛𝑑𝑒𝑥$ + 𝛽!' 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑒𝑠",%,$ 𝑆𝑡𝑟𝑖𝑛𝑔𝑒𝑛𝑐𝑦_𝑖𝑛𝑑𝑒𝑥$+ 𝛽!! 𝐻𝑜𝑠𝑝𝑖𝑡𝑎𝑙_𝑎𝑑𝑚𝑖𝑠𝑠𝑖𝑜𝑛$ + 𝛽!( 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑒𝑠",%,$ 𝐻𝑜𝑠𝑝𝑖𝑡𝑎𝑙_𝑎𝑑𝑚𝑖𝑠𝑠𝑖𝑜𝑛$ + 𝛽!)𝐷𝑒𝑎𝑡ℎ𝑠_𝑙𝑎𝑠𝑡_𝑤𝑒𝑒𝑘$ + 𝛽!* 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑒𝑠",%,$ 𝐷𝑒𝑎𝑡ℎ𝑠_𝑙𝑎𝑠𝑡_𝑤𝑒𝑒𝑘$+ 𝛽!+ 𝑆𝑒𝑎𝑟𝑐ℎ_𝑡𝑒𝑟𝑚$ + 𝛽!,𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑒𝑠",%,$ 𝑆𝑒𝑎𝑟𝑐ℎ_𝑡𝑒𝑟𝑚$+ 𝛽!- 𝐶𝑟𝑒𝑑𝑖𝑏𝑖𝑙𝑖𝑡𝑦",%,$ + 𝛽!. 𝐶𝑟𝑒𝑑𝑖𝑏𝑖𝑙𝑖𝑡𝑦",%,$ 𝑆𝑡𝑟𝑖𝑛𝑔𝑒𝑛𝑐𝑦_𝑖𝑛𝑑𝑒𝑥$ + 𝛽!/ 𝐶𝑟𝑒𝑑𝑖𝑏𝑖𝑙𝑖𝑡𝑦",%,$ 𝐻𝑜𝑠𝑝𝑖𝑡𝑎𝑙_𝑎𝑑𝑚𝑖𝑠𝑠𝑖𝑜𝑛$ + 𝛽(' 𝐶𝑟𝑒𝑑𝑖𝑏𝑖𝑙𝑖𝑡𝑦",%,$ 𝐷𝑒𝑎𝑡ℎ𝑠_𝑙𝑎𝑠𝑡_𝑤𝑒𝑒𝑘$ + 𝛽(!𝐶𝑟𝑒𝑑𝑖𝑏𝑖𝑙𝑖𝑡𝑦",%,$ 𝑆𝑒𝑎𝑟𝑐ℎ_𝑡𝑒𝑟𝑚$ + 𝛽(( 𝐶𝑜𝑛𝑡𝑎𝑐𝑡",%,$ + 𝛽() 𝐶𝑜𝑛𝑡𝑎𝑐𝑡",%,$ 𝑆𝑡𝑟𝑖𝑛𝑔𝑒𝑛𝑐𝑦_𝑖𝑛𝑑𝑒𝑥$+ 𝛽(* 𝐶𝑜𝑛𝑡𝑎𝑐𝑡",%,$ 𝐻𝑜𝑠𝑝𝑖𝑡𝑎𝑙_𝑎𝑑𝑚𝑖𝑠𝑠𝑖𝑜𝑛$ + 𝛽(+ 𝐶𝑜𝑛𝑡𝑎𝑐𝑡",%,$ 𝐷𝑒𝑎𝑡ℎ𝑠_𝑙𝑎𝑠𝑡_𝑤𝑒𝑒𝑘$+ 𝛽(,𝐶𝑜𝑛𝑡𝑎𝑐𝑡",%,$ 𝑆𝑒𝑎𝑟𝑐ℎ_𝑡𝑒𝑟𝑚$ + 𝛽(- 𝐹𝑜𝑟𝑚𝑎𝑙𝑖𝑡𝑖𝑒𝑠",%,$+ 𝛽(. 𝐹𝑜𝑟𝑚𝑎𝑙𝑖𝑡𝑖𝑒𝑠",%,$ 𝑆𝑡𝑟𝑖𝑛𝑔𝑒𝑛𝑐𝑦_𝑖𝑛𝑑𝑒𝑥$ + 𝛽(/ 𝐹𝑜𝑟𝑚𝑎𝑙𝑖𝑡𝑖𝑒𝑠",%,$ 𝐻𝑜𝑠𝑝𝑖𝑡𝑎𝑙_𝑎𝑑𝑚𝑖𝑠𝑠𝑖𝑜𝑛$ + 𝛽)' 𝐹𝑜𝑟𝑚𝑎𝑙𝑖𝑡𝑖𝑒𝑠",%,$ 𝐷𝑒𝑎𝑡ℎ𝑠_𝑙𝑎𝑠𝑡_𝑤𝑒𝑒𝑘$ + 𝛽)!𝐹𝑜𝑟𝑚𝑎𝑙𝑖𝑡𝑖𝑒𝑠",%,$ 𝑆𝑒𝑎𝑟𝑐ℎ_𝑡𝑒𝑟𝑚$ + 𝛽)( 𝑀𝑖𝑠𝑡𝑎𝑘𝑒𝑠",%,$ + 𝛽)) 𝑀𝑖𝑠𝑡𝑎𝑘𝑒𝑠",%,$ 𝑆𝑡𝑟𝑖𝑛𝑔𝑒𝑛𝑐𝑦_𝑖𝑛𝑑𝑒𝑥$+ 𝛽)* 𝑀𝑖𝑠𝑡𝑎𝑘𝑒𝑠",%,$ 𝐻𝑜𝑠𝑝𝑖𝑡𝑎𝑙_𝑎𝑑𝑚𝑖𝑠𝑠𝑖𝑜𝑛$ + 𝛽)+ 𝑀𝑖𝑠𝑡𝑎𝑘𝑒𝑠",%,$ 𝐷𝑒𝑎𝑡ℎ𝑠_𝑙𝑎𝑠𝑡_𝑤𝑒𝑒𝑘$ + 𝛽), 𝐶𝑜𝑛𝑡𝑎𝑐𝑡",%,$ 𝑆𝑒𝑎𝑟𝑐ℎ_𝑡𝑒𝑟𝑚$ + 𝛽)- 𝐼𝑛𝑣𝑜𝑙𝑣𝑒𝑚𝑒𝑛𝑡",%,$ + 𝛽). 𝐼𝑛𝑣𝑜𝑙𝑣𝑒𝑚𝑒𝑛𝑡",%,$ 𝑆𝑡𝑟𝑖𝑛𝑔𝑒𝑛𝑐𝑦_𝑖𝑛𝑑𝑒𝑥$ + 𝛽)/ 𝐼𝑛𝑣𝑜𝑙𝑣𝑒𝑚𝑒𝑛𝑡",%,$ 𝐻𝑜𝑠𝑝𝑖𝑡𝑎𝑙_𝑎𝑑𝑚𝑖𝑠𝑠𝑖𝑜𝑛$ + 𝛽*' 𝐼𝑛𝑣𝑜𝑙𝑣𝑒𝑚𝑒𝑛𝑡",%,$ 𝐷𝑒𝑎𝑡ℎ𝑠_𝑙𝑎𝑠𝑡_𝑤𝑒𝑒𝑘$+ 𝛽*! 𝐼𝑛𝑣𝑜𝑙𝑣𝑒𝑚𝑒𝑛𝑡",%,$ 𝑆𝑒𝑎𝑟𝑐ℎ_𝑡𝑒𝑟𝑚$ + 𝜖",%,$
4.3 Model validation
better than the baseline model, which only included the control variables, in predicting satisfaction, several tests was performed. The log-likelihood, Akaike information criterion (AIC) and Bayesian in-formation criterion (BIC) penalised potential overfitting. For non-nested model comparisons, it is com-mon to use information criteria such as the AIC and the BIC (Leeflang et al., 2015). The precision of the estimated model can be represented by the log-likelihood; the higher the likelihood L, the better the model fits.
5. Results
5.1 Satisfaction
Figure 8 visualises the influence of the coronavirus crisis. In the beginning of 2020, customer satisfac-tion was at an all-time high. However, this changed when the first wave of infecsatisfac-tions started at the end of February and ended in June (RIVM, 2020). As a result, overall customer satisfaction declined to below a score of 7. Due to governmental
meas-urements, hospital admissions and deaths as a result of infections declined. Temporarily, the virus was restrained, and customer satisfaction increased. Unfortunately, according to the RIVM (2020), a second wave of infections oc-curred in September and is still growing. Fig-ure 8 visualized that this negatively effets the customer satisfaction.
Figure 8 Monthly customer satisfaction 2018, 2019 and 2020.
5.2 Model 1
Table 5 Parameter estimates multilevel regression for model 1.
*** p < .001, ** p < .01, * p < .05, (.) p < .1
The various service levels had a significant impact at both the respondent and company level. Below is a brief explanation of the results of the first part of every two-part hypothesis. The results in this section start by explaining if a certain service dimension had the forecasted effect on customer satisfaction. Furthermore, for every investigated service level, the text below contains a visualization of the fluctu-ating effect on satisfaction over the past years.
Estimate (2018) Estimate (de-viation in 2019) Estimate (deviation in 2020) Hypothesised effect Fixed part Intercept 6.006 *** 0.429 0.830 * Age 0.007 ** -0.007 * -0.006 . Gender (female) 0.204 ** 0.095 -0.220 *
Income: above modal 0.109 -0.113 -0.152 Income: appr. modal -0.018 -0.086 -0.167
Income.na -0.026 0.000 -0.178 Segment discount -0.020 -0.021 -0.117 Segment service 0.056 -0.015 -0.010 H1: Employees 0.188 *** 0.021 0.101 2018: + 2020: - H2: Credibility 0.573 *** 0.059 -0.173 * 2018: + 2020: - H3: Contact 0.312 *** -0.096 -0.100 2018: + 2020: + H4: Formalities 0.338 *** 0.014 0.015 2018: + 2020: + H5: Mistakes 0.281 *** -0.169 ** 0.011 2018: + 2020: + H6: Involvement 0.466 *** 0.010 0.068 2018: + 2020: + Recommendation 0.274 *** -0.031 0.160 * Controlled for: Province √ (11 dummies) Random intercept Respondent ID 1.115 (1.056) Company 0.005 (0.072) Model fit
Null model Model 1 *** Log-likelihood -18602 -16736
AIC 37210 33630
5.2.1 Employees
Consistent with the first hypothesis, the results of Table 5 show that the availability of employees af-fected satisfaction in a significant and positive way (𝛽.= 0.188, 𝜌 < .001∗∗∗) . However,
com-paring the baseline year of 2018 and 2020 yielded an insignificant effect (𝛽!' = 0.101, 𝜌 > .05).
Figure 9 Monthly customer satisfaction with employees. This output shows that respondents thought that the first service dimension of employee availability does not differ between 2020 and the baseline year. Figure 9 shows the effect on satisfaction for this service dimension. For every service dimension, respondents could indicate whether they agreed or dis-agreed with the statement for a particular supermarket. During the first wave of infections in 2020, for both respondents who had selected “Agree” or “Not agree” on the survey question about availability of supermarket employees, there seemed to be a dip in satisfaction. Noticeably, after the first wave of infections, satisfaction quickly climbed back to approximately where it had been. Employees might not have been available, but customers might not have minded it. Customers might have been used to the situation and more forgiving after the first wave.
5.2.2 Credibility
The second hypothesis stated that the credibility of supermarkets has had a positive effect on customer satisfaction. The results of the multilevel regression in Table 5 indicate that there was a significant pos-itive effect on customer satisfaction in 2018 (𝛽.=
0.188, 𝜌 < .001∗∗∗).
Figure 10 Monthly customer satisfaction with credibility. However, in comparison with 2018, the credibility of supermarkets had a significant negative effect on customer satisfaction in 2020 (𝛽!(= −0.173, 𝜌 < .05∗ ). This implies that respondents perceived the
5.2.3 Contact
The third hypothesis forecast that if employees un-necessarily disturbed customers less, perceived satisfaction would be higher. According to Table 5, this was the case for 2018(𝛽!)= 0.312, 𝜌 < .001∗∗∗). However, for 2020, no significant
ef-fects were shown.
Figure 11 Monthly customer satisfaction with contact. This implies that the level of disturbed shoppers is not significantly different than the baseline year (𝛽!* = −0.100, 𝜌 > .05). Figure 11 shows the effect on satisfaction for this service dimension. During
the first wave of infections in 2020, for respondents who selected both “Agree” and “Not agree”, there seemed to be a dip in satisfaction concerning the amount of unnecessarily disturbed shoppers. Noticea-bly, the dip was larger for respondents who selected “Agree” than for those who selected “Not agree”. Respondents that opted “Agree”, said that employees of the supermarket in question did not unneces-sarily disturb customers. This implies that people who opted “Agree” during the first wave, seem less satisfied with this particular service dimension.
5.2.4 Formalities
Additionally, consistent with the fourth hypothe-sis, Table 5 shows that the way supermarkets han-dled formalities affected customer satisfaction in a significant and positive way (𝛽!+= 0.338, 𝜌 <
.001∗∗∗). However, comparing the baseline year
of 2018 and 2020 yielded an insignificant effect.
5.2.5 Mistakes
The fifth hypothesis suggested that supermarkets that admitted mistakes and solved problems would receive higher overall customer satisfaction scores. Table 5 shows that admitting to mistakes and solv-ing problems had a significant positive relationship with satisfaction in 2018 (𝛽!-= 0.281, 𝜌 <
.001∗∗∗).
Figure 13 Monthly customer satisfaction with mistake handling. In 2020 however, no significant effect was found. This implies that in that year, consumers did not perceive supermarkets’ handling of formalities differently than in previous years (𝛽!. = 0.011, 𝜌 > .05). Figure 13 shows the effect on satisfaction for this service dimension. During the first wave of infections in 2020, for respondents selecting both “agree” and “not agree”, there seemed to be a dip in satisfaction concerning problem solving. Noticeably, the dip for respondents who selected “agree” was larger than for those who selected “not agree”. During the first wave, people were less satisfied with the way supermarkets handled mistakes.
5.2.6 Involvement
Finally, the sixth hypothesis projected that super-markets that are more involved with their custom-ers would receive higher customer satisfaction scores. The results in table 5 shows that the level of involvement with customers had a significant positive relationship with satisfaction in 2018 (𝛽!/ = 0.466, 𝜌 < .001∗∗∗). On the other hand,
there were no significant effects for 2020.
Figure 14 Monthly customer satisfaction with involvement. This implies that in 2020, the perceived level of supermarkets’ involvement with customers was no different than in previous years (𝛽('= 0.068, 𝜌 > .05). Figure 14 shows the effect on satisfaction for
5.3 Supermarket heterogeneity
Regarding the supermarket segments (Table 6), no significant effect highlighted differences across su-permarket segments over the study period. Specifically, for the grocery stores within the discount seg-ment, the study found a non-significant negative effect on satisfaction (𝛽+= −0.020, 𝜌 > .05). For
supermarkets within the service segment, there was a non-significant positive effect on satisfaction (𝛽+= 0.056, 𝜌 > .05). Table 6 contains the customer satisfaction scores for every supermarket chain individually, over the past three years.
Table 6 Satisfaction per brand at the customer level.
Satisfaction 2018 2019 2020 Service segment Albert Heijn 7.69 7.73 7.91 Jumbo 7.82 7.68 7.72 Picnic 8.21 7.37 8.72 Plus 7.91 7.89 7.69 Poiesz 7.58 6.97 7.71 Spar 7.22 7.61 7.57 Middle segment Coop 7.52 7.66 7.48 Deen 7.65 7.85 7.93 DekaMarkt 7.32 7.23 7.58 Hoogvliet 7.51 7.71 7.85 Jan Linders 7.98 8.01 7.97 Makro 7.12 7.09 7.15 Discount segment Aldi 7.25 7.26 7.40 Boni 7.58 7.89 7.68 Dirk v/d Broek 7.68 7.57 7.49 Lidl 7.55 7.46 7.61 Nettorama 7.43 7.53 7.40 Vomar 6.62 7.47 7.15
5.4 Model 2
stronger in 2020 during the Corona crisis. A second model was designed to interpret interactions
with the four coronavirus variables for every service dimension (Equation 2). The variables included to measure the effects of the pandemic were stringency index, hospital admissions, deaths last week andGoogle search terms. This model includes 712.469 observations distributed over the three supermarkets
segment. The “Discount” segment contained 251.079 observations, the “Middle” segment 101.345 ob-servations and the “Service” segment 360.045 obob-servations. Table 7 shows that Model 2 is highly sig-nificant and shows better log-likelihoods, AIC and BIC scores in comparison with the null model. Table 7 Parameter estimates multilevel regression for model 2.
Estimate 2020 Stringency index Hospital admissions Deaths last week Search term Hypoth. effect Fixed part Intercept 6.264 *** 0.001 0.005 -0.030 *** 0.002 Age 0.000 Gender (female) -0.078 Income: above modal -0.138 Income: appr. modal -0.234 . Income.na -0.210 Segment discount -0.137 Segment service 0.257 . H1: Employees 0.519 *** -0.042 *** -0.003 0.047 *** -0.017 *** Main: + Interaction: - H2: Credibility 0.390 *** 0.016 *** -0.022 *** 0.002 *** -0.014 *** Main: + Interaction: - H3: Contact 0.375 *** 0.001 0.000 -0.004 *** -0.005 *** Main: + Interaction: + H4: Formalities 0.023 . -0.035 *** 0.001 -0.012 *** 0.011 *** Main: + Interaction: + H5: Mistakes 0.044 *** 0.005 *** 0.000 -0.002 *** 0.017 *** Main: + Interaction: + H6: Involvement 0.315 *** 0.012 *** 0.020 *** 0.010 *** 0.002 *** Main: + Interaction: + Recommendation 0.373 *** Controlled for: Province √ (11 dummies) Random intercept Respondent ID 1.496 (1.223) Company 0.060 (0.245) Model fit
Null model Model 2 *** Log-likelihood -870664 -655532
5.4.1 Employees
Consistent with the first hypothesis, the results of Table 7 show that the availability of employees af-fected satisfaction in a significant and positive way in 2020 (𝛽. = 0.519, 𝜌 < .001∗∗∗). Four variables
were integrated to understand the effects of the pandemic on the customer satisfaction. The effects will be measured by looking at the effect of the interactions between the pandemic related variables and every service dimension. The first interaction effect of the stringency index shows a significant negative effect on employees (𝛽!'= −0.042, 𝜌 < .001∗∗∗). The second interaction effect of hospital
admis-sions indicates a non-significant effect on employees (𝛽!(= −0.003, 𝜌 > .05). The third interaction effect of deaths last week signifies a significant positive effect on employees (𝛽!*= 0.047, 𝜌 < .001∗∗∗). The last interaction effect of the mean of the search terms shows a significant negative effect
on employees (𝛽!, = −0.017, 𝜌 < .001∗∗∗). Thus, in line with the hypothesis (1a), Table 7 suggests a
positive effect for the availability of employees during the COVID-19 crisis on customer satisfaction. However, both negative and positive interaction effects were found among the pandemic related varia-bles and the availability of employees. The second part of the first hypothesis (1b) suggested a negative effect on the relationship between the availability of employees and customer satisfaction. This hypoth-esis can be partly accepted due to the partly negative interaction effects between the Corona related variables and the availability of employees.
5.4.2 Credibility
The results of the multilevel regression in Table 7 point to a significant positive main effect of the credibility of supermarkets on customer satisfaction in 2020 (𝛽!-= 0.390, 𝜌 < .001∗∗∗). This finding
is in line with the second hypothesis. The first inter-action effect of the stringency index shows a signif-icant positive effect on credibility (𝛽!.= 0.016, 𝜌 < .001∗∗∗). The second interaction effect
of hospital admissions also shows a significant pos-itive effect on credibility (𝛽!/ = −0.022, 𝜌 <
.001∗∗∗).
Figure 15 Relationship between deaths in the previous week and the stringency index.
The third interaction effect of deaths last week points to a significant positive effect on credibility (𝛽('= 0.002, 𝜌 < .001∗∗∗). Finally, the last interaction effect of the mean of the search terms indicates
a significant negative effect on credibility (𝛽(!= −0.014, 𝜌 < .001∗∗∗).
To conclude, in line with the hypothesis (2a), the main effect in Table 7 suggests a positive effect for the credibility of supermarkets on customer satisfaction in 2020. The second part of the hypothesis (2b) suggested a negative effect on the relationship between the credibility of supermarkets and customer satisfaction due to the Corona included variables. However, both negative and positive interaction ef-fects were demonstrated by the integrated Corona related variables. This hypothesis was partly accepted due to the partly negative interaction effects.
5.4.3 Contact
The multilevel regression in Table 7 suggested a significant positive main effect on customer satisfaction in 2020 (𝛽((= 0.375, 𝜌 < .001∗∗∗). The first interaction effect of the stringency index shows an
in-significant effect on contact (𝛽()= 0.001, 𝜌 > .05). The second interaction effect of hospital
admis-sions indicates an insignificant effect on contact as well (𝛽(*= 0.000, 𝜌 > .05). The third interaction
effect of deaths last week signifies a significant negative effect on contact (𝛽(+ = −0.004, 𝜌 <
.001∗∗∗). Finally, the last interaction effect of the mean of the search terms shows a significant negative
effect on contact (𝛽(,= −0.005, 𝜌 < .001∗∗∗).
Therefore, in line with the hypothesis (3a), the main effect in Table 7 points to a positive effect for contact on customer satisfaction in 2020. The second part of the hypothesis (3b) forecasted a positive effect on the relationship between the service dimension contact and customer satisfaction during the pandemic. This implied that the level of customers that are not unnecessarily disturbed while shopping would indicate higher perceived satisfaction during the pandemic. However, both variables that signif-icantly affected contact had negative coefficients. This research suggests that during the coronavirus crisis, the level of customers that are not unnecessarily disturbed while shopping is negatively affected and results in a lower customer satisfaction.
5.4.4 Formalities
The multilevel regression in Table 7 points to a marginally significant main effect of formalities on customer satisfaction in 2020 (𝛽(-= 0.023, 𝜌 < .1 ). This implies that in 2020, the way supermarkets
handled formalities, such as the possibility to return goods, marginally and significantly affected satis-faction. Next to the main effect, there were also crossover interactions. The first interaction effect of the stringency index shows a significant negative effect on formalities (𝛽(.= −0.035, 𝜌 < .001∗∗∗). The
second interaction effect of hospital admissions indicates an insignificant effect on formalities (𝛽(/ =
0.001, 𝜌 > .05). The third interaction effect of deaths last week suggests a significant negative effect on formalities (𝛽)'= −0.012, 𝜌 < .001∗∗∗). Finally, the last interaction effect of the mean of the
search terms signifies a significant positive effect on formalities (𝛽)!= 0.011, 𝜌 < .001∗∗∗). Thus, in
5.4.5 Mistakes
The multilevel regression in Table 7 revealed a significant positive main effect of supermarkets that admitted mistakes and solved problems on customer satisfaction in 2020 (𝛽)(= 0.044, 𝜌 < .001∗∗∗).
The first interaction effect of the stringency index shows a significant positive effect on mistakes admit-ted and problems solved (𝛽))= 0.005, 𝜌 < .001∗∗∗). The second interaction effect of hospital
admis-sions indicates a non-significant effect on mistakes as well (𝛽)*= 0.000, 𝜌 > .05). The third interac-tion effect of deaths last week shows a significant negative effect on mistakes (𝛽)+ =
−0.002, 𝜌 < .001∗∗∗). Finally, the last interaction effect of the mean of the search terms suggests a
significant positive effect on mistakes (𝛽),= 0.017, 𝜌 < .001∗∗∗). Thus, in line with the hypothesis
(5a), the main effect in Table 7 points to a positive effect for supermarkets’ problem solving during the COVID-19 crisis. The second part of the hypothesis (5b) suggested a positive effect on the relationship between supermarkets that admitted mistakes and solved problems and customer satisfaction due to COVID-19. However, both negative and positive effects were demonstrated by the integrated variables. This hypothesis was partly accepted due to the partly negative interaction effects.
5.4.6 Involvement
The multilevel regression in Table 7 suggested a significant positive main effect of the level of involve-ment on customer satisfaction in 2020 (𝛽)-= 0.315, 𝜌 < .001∗∗∗). The first interaction effect of the
stringency index shows a significant positive effect on the level of involvement with customers (𝛽). = 0.012, 𝜌 < .001∗∗∗). The second interaction effect of hospital admissions indicates a significant
posi-tive effect on the level of involvement as well (𝛽)/ = 0.020, 𝜌 < .001∗∗∗). The third interaction effect
of deaths last week signifies a significant positive effect on involvement (𝛽*'= 0.010, 𝜌 <
.001∗∗∗). Finally, the last interaction effect of the
mean of the search terms shows a significant pos-itive effect on involvement (𝛽*!= 0.002, 𝜌 < .001∗∗∗).
Figure 16 Relationship between hospital admissions and the stringency index. Figure 16 contains a comparison of the stringency index and hospital admissions due to the coronavirus. Both variables had a significant effect on the level of involvement of supermarkets in 2020.
involvement and had a positive slope. During the crisis, the involvement service dimension has been positively affected by all integrated pandemic variables.
6. Discussion and conclusion
In marketing research, many scholars have focused on drivers of customer satisfaction and evaluating crises of the past. Firm performance is the variable most often cited as the main driver in the recent literature. This research, however, has focused on customer perceptions during a crisis. The pandemic has caused marketers to adapt their strategies and contribute to tackling crisis-related problems. These “purpose marketing” activities are the new benchmark of 2020 (Hoekstra & Leeflang, 2020).
This study offers more insight into the impact of a crisis on perceived service quality and, ultimately, customer satisfaction. In addition, by analysing a large number of supermarkets, this study can offer managerial guidance on how the impact of the pandemic has differed for grocery stores. The multilevel regression model resulted in new insights in this area. First, the results show that the customer satisfac-tion was at an all-time high in the beginning of 2020. However, this changed when the first wave of infections started at the end of February and ended in June (RIVM, 2020). To investigate this effect further, a multilevel regression model was estimated that showed the differences over the years. The results indicated that all service dimensions significantly and positively affected customer satisfaction. More interestingly, regarding the interaction effects of the year 2020, only the service dimension of credibility seemed to be significantly different in 2020 than in 2018. This implies that respondents per-ceived supermarkets’ level of credibility as different in this year. To investigate if the current pandemic had played a role, the second multilevel regression model was developed. The variables used to measure the impact of the coronavirus yielded results that differed across the service dimensions. As a result, the formulated hypotheses could be rejected or accepted. The first hypothesis was formulated as follows: “The positive impact of the availability of supermarket employees on customer satisfaction (1a) has been negatively affected by the coronavirus crisis (1b).” The findings resulted in hypothesis 1a being accepted and hypothesis 1b being partly accepted due to both positive and negative interaction effects. This outcome implies that the availability of supermarket employees has a positive effect on customer satisfaction and that this relationship has been partly affected by the coronavirus crisis.
The third hypothesis was worded as follows: “The positive impact of customers not being unnecessarily disturbed by supermarket employees while shopping on customer satisfaction (3a) has been strength-ened by the coronavirus crisis (3b).” Based on the results, the first part of the hypothesis (3a) was ac-cepted, and the second part (3b) was rejected. Due to the fact that both included Corona variables were significant and had a negative effect on the relationship between customers not being unnecessarily disturbed while shopping and customer satisfaction. The results suggest that during the coronavirus cri-sis, the level of customers that are not unnecessarily disturbed while shopping is negatively affected and results in a lower customer satisfaction. This would imply that people are unnecessarily disturbed more during shopping, which seems odd with social distance measures. It could be however, that shoppers are more conscious of their surroundings what could result in irritations when too crowded.
The fourth hypothesis concerned the handling of formalities and read as follows: “The positive impact of the way that customer’s experience handled formalities by supermarkets on customer satisfaction (4a) has been positively affected by the coronavirus crisis (4b).” Hypothesis 4a is accepted due to the mar-ginal significance of the main effect, and hypothesis 4b was partly accepted due to the positive and negative interaction effects.
The fifth hypothesis was formulated as follows: “The positive impact of supermarkets’ problem solving on customer satisfaction (5a) has been positively affected by the coronavirus crisis (5b).” Based on the results, hypothesis 5a was accepted, and hypothesis 5b was partly accepted due to the presence of both positive and negative interaction effects. This outcome implies that supermarkets’ complaint handling has a positive effect on customer satisfaction and that this relationship has been partly affected by the coronavirus crisis.
The sixth hypothesis concerned the level of involvement with customers: “The positive impact of su-permarkets’ involvement with their customers on customer satisfaction (6a) has been positively affected by the coronavirus crisis (6b).” The results meant that hypotheses 6a and 6b were accepted. This out-come implies that the level of customer involvement has a positive effect on customer satisfaction and that this relationship has been affected by the coronavirus crisis.
