key: cord-0065405-iuhxtg4q
authors: Wang, Yichuan; Zhang, Minhao; Li, Shuyang; McLeay, Fraser; Gupta, Suraksha
title: Corporate Responses to the Coronavirus Crisis and their Impact on Electronic‐Word‐of‐Mouth and Trust Recovery: Evidence from Social Media
date: 2021-03-30
journal: nan
DOI: 10.1111/1467-8551.12497
sha: 22d7a5d59cbb4d169eb65d4131c48e19278e4797
doc_id: 65405
cord_uid: iuhxtg4q

This study examines how corporate responses to service failure, caused by the coronavirus (COVID‐19) crisis, influence electronic‐word‐of‐mouth (E‐WoM) and trust recovery around lockdown, using multiple data sources. A dataset of 398 valid COVID‐19 announcements from 50 UK food retailers posted on the social media platform Twitter, and 21,960 consumer comments associated with these announcements, are analysed using content analysis and social media analytics, respectively. In Study 1, we test the effects of corporate crisis response strategy (defensive vs. offensive) and response framing (emotional vs. rational) on consumer E‐WoM (measured as ‘consumer sentiment’). The results reveal that using a defensive corporate response strategy with emotionally framed announcements leads to more positive consumer E‐WoM. In Study 2, we advance the findings of Study 1 using a vignette‐based experimental design to examine how social media announcements made by food retailing brands influence consumers’ trust recovery. We find that consumer trust recovers significantly when corporate COVID‐19 responses are framed in an emotional manner. By drawing upon signalling theory, this study makes an important contribution to public health crisis communication and service failure literature by demystifying consumers’ reactions towards corporate crisis responses amid a pandemic.

The coronavirus disease, officially termed COVID-19 by the World Health Organization (WHO), has been viewed as a public health emergency. 1 The 1 According to the WHO's dictionary, a public health emergency is defined as: 'An occurrence or imminent threat of an illness or health condition, caused by bio terrorism, epidemic or pandemic disease, or (a) novel and highly fatal infectious agent or biological toxin, that poses a substantial risk of a significant number of human fatalities or incidents or permanent or long-term disability.' emerging COVID-19 crisis has had a dramatic effect on many business sectors and may trigger a recession (Goodman, 2020) . At the current time, the ultimate impact of the pandemic remains ambiguous, as vaccines are starting to be approved and slowly administered. However, many governments have had to introduce, and subsequently reintroduce, measures such as lockdowns, social distancing and strict hygiene requirements to respond to the pandemic crisis. This crisis has undoubtedly

The definition of public health emergency can be found at https://www.who.int/hac/about/definitions/en/ 2 Y. Wang et al. resulted in service failure, 2 as many retailers have closed, run out of stock or had to ration purchases, with consumers having to queue to enter supermarkets during the initial lockdown. Consumers have changed their purchasing patterns and behaviours, and this has led to more online shopping, stockpiling and bulk-buying (He and Harris, 2020) .

We build upon the public health crisis communication and service failure literature as well as signalling theory to explore the influences which corporate crisis communication responses have on consumers' behavioural reactions in social media environments. By developing and testing a theoretical framework which demonstrates the interaction effects that corporate crisis response strategy (offensive vs. defensive strategy) and response framing (emotional vs. rational) have on consumer electronic-word-of-mouth (E-WoM) and trust recovery, this paper contributes to the development of signalling theory in the context of corporate responses to a public health crisis. In doing so, we answer recent calls for further research which focuses on service failure and recovery in a new context (i.e. a public health crisis), takes a multi-source perspective and uses big data from social media (Grégoire and Mattila, 2020; Khamitov, Grégoire and Suri, 2020) . This study seeks to answer the following three research questions: RQ1: How do corporate social media announcements relating to a public health crisis influence consumer E-WoM and trust recovery? RQ2: During a public health crisis, what response strategy (offensive vs. defensive) and response framing (emotional vs. rational) on social media is most effective in driving consumer positive E-WoM and trust recovery? RQ3: How do the effects of crisis response strategy and crisis response framing interact in driving consumer positive E-WoM and trust recovery?

To answer our research questions, two studies are conducted. Study 1 examines UK food retailers' announcements relating to COVID-19 posted on Twitter over a 2-month period spanning 1 March 2020 (3 weeks before lockdown measures were announced in the UK) through to 30 April 2020 (the day when it was announced that lockdown measures would be reduced from 10 May) (BBC, 2020) . A dataset comprising 50 UK food retailers, with 398 valid, corporate COVID-19 crisis communication announcements posted on Twitter, as well as 21,960 consumer comments (replies) related to these announcements, is developed and analysed. Study 2 examines changes in consumer trust towards different corporate crisis response strategies and framing approaches during the pandemic. A three-stage experiment is designed to consider one UK food retailer and two different pandemic crisis vignettes, which create eight mock-Twitter tweets.

According to the crisis management literature, developing appropriate corporate response strategies is key to addressing a public health crisis (Malet and Korbitz, 2015; Page et al., 2006; Volo, 2008) . During a crisis, response is defined as the ability to react to emerging risks through reconfiguring resources and to implement appropriate crisis communication strategies (Ritchie, 2008) . In Appendix A, we summarize the representative empirical studies focusing on public health crisis responses. Although crisis management studies provide examples of well-documented experiences and best practice to combat a crisis, we identify several research gaps which should be addressed.

First, the extant research has frequently discussed the important role of governmental communication in addressing the challenges created by a public health crisis. An appropriate governmental risk communication strategy will enable the business sector to recover more quickly from a public health crisis (e.g. Malet and Korbitz, 2015; Page et al., 2006; Volo, 2008) . While social media has been widely regarded as an important communication platform, the existing literature has not specifically focused on explaining how firms should use social media to communicate with their customers during a public health crisis.

Second, the severity and duration of the COVID-19 pandemic is unprecedented compared to all other types of recent crisis. Its reach is global, and the disease is highly contagious compared to other epidemic, or pandemic, disease outbreaks such as SARS, H1N1, MERS, Ebola and Zika. Therefore, it is necessary to revisit and update the existing models, strategies and theories of public health crisis communication to cope with the uniqueness, complexity and unpredictability of this global pandemic (Knowles et al., 2020) .

Third, the government sector has been a particular focus of a wide range of relevant studies (e.g. Chong, 2006; Novelli et al., 2018) , attributed to the fact that government authorities play a primary role in public health crisis communication. However, the retail and consumption-related industries have received much less attention. UK food retailers, such as supermarkets, have been a business sector hugely impacted by the COVID-19 outbreak (TextAnywhere, 2020). For example, the UK's vegetable box providers were expected to deliver 3.5 million boxes to their customers within 2 months during the summer of 2020; double the demand of the pre-COVID-19 period (Food Foundation, 2020) . In addition, consumer stockpiling and panic buying behaviour triggered a shortage of essentials and a growing distrust of UK food retailers' ability to cope with unprecedented demand (Connors et al., 2020) . Due to the demand uncertainty and consumers' concerns about their safety while shopping, corporations have suffered from service failures and a loss of customer trust (Lee et al., 2020) . Therefore, the establishment of effective crisis communication and service recovery strategies is an urgent priority for UK food retailers, seeking to prioritize customer needs and recover customer trust (Broadbent, 2020; Bunker, 2020) .

Fourth, E-WoM is increasingly being used by consumers to express their displeasure with service failure (Fouroudi et al., 2020) . E-WoM involves positive, negative or neutral statements that consumers make about a product, or a corporation, which are available for others to view on the Internet (Filieri and McLeay, 2014) . We envision that managers attempt to learn from consumergenerated E-WoM as they attempt to evaluate their corporate responses to the pandemic. In particular, E-WoM may provide evidence of a recovery from a loss of consumer trust due to service failure. Prior research has often overlooked or ignored the rich social media data generated by consumers in service failure contexts. We address this gap by using social media data to better understand the consumer experience during service failures caused by COVID-19.

To fill the gaps in prior research, we draw upon signalling theory to provide a theoretical lens for our conceptual model and analysis. Signalling theory is concerned with building communication relationships and reducing information asymmetries between two parties (a sender and a receiver) (Connelly et al., 2011; Walsh et al., 2009) . Signals are considered as the presentation of corporate announcements and can convey important information regarding the sender's intention and abilities (Musteen, Datta and Kemmerer, 2010) . For instance, when corporations experience atypical changes in their service provision (service failure), they often communicate, or signal, the changes to consumers who may interpret such information in different manners. Signalling theory has frequently been used in the service failure literature to study approaches to relationship recovery (Kharouf et al., 2020) , to explain the effectiveness of corporate announcements (Ni, Flynn and Jacobs, 2016) and to examine product recall communications (Hsu and Lawrence, 2016) . As such, the application of signalling theory has proven useful in advancing our understanding of where corporate signals originate, what specific signals flow from those sources and the resultant impact these signals may have on the behavioural reactions of customers (Drover, Wood and Corbett, 2018) .

We apply signalling theory in an attempt to better understand how consumers react to corporate social media announcements relating to service failure caused by the global pandemic crisis. The literature suggests that during a crisis, consumers are likely to pay close attention to corporations' response strategies on social media, especially actions that reduce the negative impact of the crisis on consumers (Tse et al., 2016) . Coombs (2007) describes crisis response strategy as an organization's answer to an unexpected event in the hope of repairing reputation, reducing negative impacts and preventing negative behavioural intentions. Crisis response framing, defined by Claeys and Cauberghe (2014) , refers to the use of language to convey messages which appeal to individual feelings and have an effect on consumer behaviour. The strategy and communication frame which signallers adopt aims to intentionally communicate information, thereby conveying the positive attributes of a corporation as they attempt to obtain a good consumer reputation (Connelly et al., 2011; Walsh et al., 2016) . Therefore, our research model aims to investigate the varied effects of crisis response (i.e. response framing and response strategy signals) on consumer reactions to corporate service failure. The research model contains two studies and is illustrated in Figure 1 .

In Study 1, we examine the effects of corporate response strategies and framing on customers' E-WoM by analysing a large volume of consumer responses, which enhances the external validity and supports the generalizability of the study. In Study 2, we investigate how, and if, corporate responses enable consumers to recover trust using a scenario-based experiment, establishing internal validity based on precisely controlling and measuring changes in consumer trust. In a crisis situation, understanding E-WoM and enhancing trust recovery are two critical areas of concern for managers. However, they are also two distinctive concepts which are difficult to measure through a single data source, or examine using a single method. Although the management and marketing literature has widely accepted the use of social media data to measure E-WoM, customer trust is a highly abstract concept which is difficult to measure using social media data alone (Pavlou and Dimoka, 2006) . Therefore, Study 2 supplements the results of Study 1.

Corporations have developed crisis response strategies to communicate with consumers during various crises, in an attempt to maintain sales, rebuild brand reputation and improve customer relationships (He, Rui and Whinston, 2018; Sela, Wheeler and Sarial-Abi, 2012) . Two opposing types of response strategies explicitly feature in the marketing literature: offensive responses and defensive responses (Fornell and Wernerfelt, 1987; Hauser and Shugan, 2008 ). An offensive corporate response strategy primarily focuses on corporations attracting new consumers or competitors' dissatisfied consumers, increasing purchases and obtaining a greater market share (Fornell and Wernerfelt, 1987; He, Rui and Whinston, 2018) . In contrast, a defensive strategy protects a corporation's market share from attack by external factors such as competition and crisis (Woodall, 2004) , with marketing efforts directed at existing customers and customer retention activities (Martin-Herran and Sigue, 2019; Roberts, 2005) . Real examples of offensive and defensive COVID-19 corporate crisis response announcements are presented in Appendix B.

During a crisis, a defensive announcement (signal) explaining a corporation's situation is likely to lead to consumers gaining a better understanding of the business and reducing the proportion of dissatisfied customers who would otherwise complain (He, Rui and Whinston, 2018) . Consumers may view a defensive announcement as proof of a corporation's determination and efforts to cope with the crisis. Drawing on signalling theory, for example, Kharouf et al. (2020) suggest that announcements focusing on retaining existing customers could repair corporate-consumer relationships and result in positive corporate reputations during a crisis. In times of COVID-19, lockdown-induced panic buying led to food retailers imposing purchasing restrictions and limiting operational hours. In response, some food retailers develop defensive communications that provide corrective and restorative guidance (e.g. frontline hygiene and social distancing restrictions) in order to explain the crisis situation to consumers. Signalling theory indicates that a receiver's perceptions of a signal are largely dependent on the extent to which signallers are honest about a situation (Connelly et al., 2011) . Hence, a proper defensive response strategy enables consumers to gain a better understanding of corporate COVID-19 contingency plans. These actions are important to reduce the proportion of dissatisfied consumers who would otherwise complain and to ensure that consumers can disseminate positive E-WoM to support food retailers, if appropriate.

To the contrary, an offensive response strategy has been acknowledged to improve customers' E-WoM and satisfaction in certain circumstances (Gao et al., 2017; Karakaya and Yannopoulos, 2010) . However, in the context of COVID-19, offensive signals which mainly emphasize the promotion of products and brands, without responding to consumers' concerns about the pandemic in an empathetic manner, are likely to cause dissatisfaction and the spread of negative E-WoM (Sheth, 2020) . Therefore, we argue that the clarification of a corporation's situation using a defensive strategy is more likely to enhance existing customers' understanding, acceptance and positive E-WoM than an offensive strategy which seeks to increase purchases or recruit new consumers. We propose the following hypothesis:

H1: A defensive crisis response strategy will result in more positive consumer E-WoM than an offensive crisis response strategy.

In addition to corporate response strategies, the ways in which announcements are framed and presented are also likely to have an impact on how consumers evaluate communication announcements and spread E-WoM during a crisis (Claeys and Cauberghe, 2014) . Consumers who receive signals make decisions based on their per-ceptions of the information presented in the signal, which can be influenced by how a message is framed (Connelly et al., 2011) . Prior studies have examined the effects of language framing by exploring the influence which emotional and rational corporate responses have on consumer behaviour (Sela, Wheeler and Sarial-Abi, 2012) . Emotional responses appeal to the use of subjective and evaluative properties in communication messages to persuade consumers, while rational responses focus on informative and straightforward messages (Claeys and Cauberghe, 2014; Orth, Koenig and Firbasova, 2007) .

Research on consumer-brand relationships has proved that high levels of (positive) emotional responses are associated with more favourable consumer perceptions of a brand and/or corporation (Frank et al., 2014; Schoefer and Diamantopoulos, 2009 ). Applied in the COVID-19 situation, where consumers generally and continuously seek support, emotional announcements indicating care and connection are critical for creating positive experiences. Moreover, the usage of emotional terms such as 'we' and 'us' in framing a message can indicate a close relationship and common identity shared between the signaller and the receiver (Sela, Wheeler and Sarial-Abi, 2012 ). E-WoM is, to a large extent, driven by the emotions that consumers perceive and feel when interacting with a corporation (Verhagen, Nauta and Feldberg, 2013) . Following these theoretical arguments, an emotionally framed, corporate crisis response will reinforce consumers' perceptions that they have a positive and supportive relationship with a corporation, leading to positive consumer E-WoM.

Additionally, from a psychological perspective, rational framing is more likely to induce cognitive reactions when consumers view a corporate announcement (Claeys and Cauberghe, 2014) . In a pandemic crisis, when consumers are separated and experience social isolation, consumers who receive signals which are framed in an emotional and supportive manner are likely to experience enhanced relationships with a corporation (Odekerken-Schroder et al., 2020) . Hence, we propose the following hypothesis:

H2: Emotionally framed responses will result in more positive consumer E-WoM than rationally framed responses. 6 Y. Wang et al. Emotional framing can also serve as a means of reinforcing customers' perceptions of the impact of corporate crisis response strategies. Prior studies in customer communications indicate that when corporations provide information, consumers are sensitive to minor changes in wording and subtle linguistic components (i.e. the incorporation of emotional or rational components) in communication signals, which, in turn, has an impact on customers' opinions of a corporation (Sela, Wheeler and Sarial-Abi, 2012) . The effect of communication strategy on consumers' behavioural actions may vary depending on how communication content is framed (Claeys and Cauberghe, 2014; Claeys, Cauberghe and Leysen, 2013; Cummings and Yule, 2020) .

The role of emotional framing cannot be neglected when developing crisis response strategies (Coombs and Holladay, 2005) . The moderation role of response framing is based on the premise that corporations will first decide on which response strategy to adopt in a crisis communication announcement, and subsequently develop the language framing. Signalling theory emphasizes the importance of consistency between a signal's content and its framing (Connelly et al., 2011) . In the crisis management context, the use of emotional framing can be an effective addition to a defensive crisis response strategy, as consumers tend to build a better understanding of the corporation's situation (Coombs, 2007) . Hence, we propose the following hypothesis:

H3a: A defensive crisis response strategy framed in an emotional manner will result in more positive consumer E-WoM than one framed in a rational manner.

We expect that the effects of language framing will be different when a corporation engages an offensive crisis response strategy which seeks to compete within the marketplace and attract new customers through price promotion or product and brand marketing (He, Rui and Whinston, 2018) . A consensus in the signalling literature is that visible, credible and structured signals are helpful in indicating a signaller's unobservable qualities and capabilities (Xia et al., 2016) . Following this line of thought, corporations can develop offensive crisis communication announcements in a rational and structured manner, which demonstrates their capability to provide better products/services than their competitors and therefore, gain a positive consumer reputation. For instance, logical messages that highlight in-store safety enhancements, or the development of competitive online delivery services, may make consumers feel safe enough to shop with them and subsequently, promote their brand using E-WoM. Hence, we expect that framing corporate crisis responses in a rational manner will strengthen the impact of an offensive response strategy and generate positive E-WoM, which leads to the following hypothesis:

H3b: An offensive crisis response strategy framed in a rational manner will result in more positive consumer E-WoM than one framed in an emotional manner.

During a crisis, attempts to regain consumers' beliefs, intentions and trust should be at the top of the strategic agenda of any corporation which has experienced service failure (Fouroudi et al., 2020; Kim et al., 2004) . As trust is often lost as a result of a service failure, it is important for managers to rectify, amend and restore losses in trust experienced by customers (Hess, Ganesan and Klein, 2003; Kharouf et al., 2020) . Trust in a brand is mainly shaped by consumers' confidence in a corporation's reliability and integrity (Morgan and Hunt, 1994) . Corporate crisis response announcements, which explain the reasons for a service failure and signal their reliability and competence at addressing negative customer experiences, tend to rebuild trust with existing customers. Rather than using an offensive strategy which draws customer attention away from the crisis, a defensive response strategy can help to build reputation and trust with existing customers as well as the wider marketplace (Luo and Homburg, 2007) .

Furthermore, defensive responses can signal a corporation's ethical, moral and fair-thinking characteristics, which consequently increase trustworthiness (Basoglu and Hess, 2014) . Similar arguments in the situational crisis and the service failure communication literature suggest that no matter how a corporation responds to negative events associated with a crisis, direct communications to customers are essential for recovering consumer trust and enabling 'forgiveness' (Coombs, 2007; Kharouf et al., 2020; Xiao, Cauberghe and Hudders, 2020) . Therefore, we propose the following hypothesis:

H4: A defensive crisis response strategy will result in greater consumer trust recovery than an offensive response strategy.

The extent to which corporations are sincere in their crisis communication announcements signals their level of consideration in consumer interest and well-being, when developing trust recovery initiatives (Xiao et al., 2018; Xie and Peng, 2009 ). In comparison with rational framing, an emotionally framed announcement, which focuses on subjective and emotional factors, is normally perceived by customers as more sincere (Claeys, Cauberghe and Leysen, 2013) . Existing research has provided evidence that messages expressing emotions when communicating a crisis response can lead to higher levels of consumer trust restoration (van der Meer and Verhoeven, 2014). For example, research in the social media communication field reveals that a crisis response message which expresses hope (i.e. emotional framing) is more effective in developing post-crisis information acceptance than a rationally framed message (Xiao, Cauberghe and Hudders, 2020) . Huang and DiStaso (2020) provide similar results in the context of highly contagious disease outbreak: an emotional appeal leads to responses being perceived as more credible, therefore increasing public trust towards a corporation.

One reason why a message framed in an emotional manner results in greater trust recovery is because it allows a corporation to express its 'human' side (McCorkindale and DiStaso, 2014) . The COVID-19 pandemic has been a nightmare for food retailers seeking to manage customer complaints about, for example, product shortages. The COVID-19 crisis is likely to have evoked negative emotions such as outrage, or dissatisfaction, among some customers and damaged trust (Malecki, Keating and Safdar, 2020) . The utilization of an emotionally framed approach will enable a corporation to express remorse for negative consequences, assert its strong willingness to take responsibility and to demonstrate hope for the future (Malecki, Keating and Safdar, 2020; Xie and Peng, 2009 ). Consumers may perceive that a corporation which responds with an emotionally framed message is seeking to solve problems and recover trustworthiness. In summary, an emotion-ally framed response implies a corporation's 'good faith' in taking responsibility for negative events. Therefore, the following hypothesis is posited:

H5: Emotionally framed responses will result in greater consumer trust recovery than rationally framed responses.

To develop effective corporate crisis communication signals, it is important to consider how the effects of the interaction between corporate response strategy and framing influence consumer trust recovery (e.g. providing the 'right' response framing with the 'right' response strategy). Prior research argues that the framing of corporate crisis communication responses can reinforce the link between a corporate response strategy and consumer reactions during, and after, a crisis (Claeys and Cauberghe, 2014; Claeys, Cauberghe and Vyncke, 2010) . For example, the results of research conducted by Claeys and Cauberghe (2014) suggest that when corporate communication messages are framed in a rational manner, consumers develop positive attitudes.

COVID-19 is categorized as a 'victim cluster' crisis (Coombs, 2007) , in which corporations are perceived as the victim rather than the cause of a crisis. Thus, we argue that defensive corporate communication strategies which are framed in an emotional manner by corporations may evoke empathy and mitigate market uncertainty and decrease consumer anxiety (Bunker, 2020; Novelli et al., 2018) and thus generate more positive E-WoM. For example, during the COVID-19 crisis, to rebuild consumer trust, Tesco introduced the 'Keeping you Safe' campaign. This campaign not only aims to provide clear guidance on social distancing, but also conveys messages associated with positive emotions (e.g. 'Together, we can do this' and '#EveryLittleHelps') to reduce consumers' anxiety and depression.

While the success of a defensive corporate crisis response strategy is likely to depend on engaging customers by appealing to their emotions, we propose that an effective offensive response strategy will be highly reliant on leveraging rational frames to recover consumer trust. Indeed, the success of an offensive response strategy is determined by whether consumers are persuaded by the offers, or value proposition, made by a corporation. The results of research undertaken by Claeys and Cauberghe (2014) suggest that framing corporate 8 Y. Wang et al. crisis response messages in a rational manner, in an effort to respond to service failure, can persuade customers to engage with a brand and develop more positive attitudes. For instance, in the UK, Aldi developed offensive response strategies which featured popular product lines 3 and emphasized the high levels of availability of the products they sell (without introducing product limits) relative to other food retailers (Kollewe and Butler, 2020) . These offensive responses, which were framed in an informative and persuasive manner, have the potential to amplify consumer trust. We therefore develop the following two hypotheses:

H6a: A defensive crisis response strategy framed in an emotional manner will result in a higher likelihood of consumer trust recovery than one framed in a rational manner. H6b: An offensive crisis response strategy framed in a rational manner will result in a higher likelihood of consumer trust recovery than one framed in a rational manner.

In Study 1, we examine: (1) the effects that corporate crisis response strategies have on E-WoM;

(2) the effects that corporate crisis response framing have on E-WoM; (3) the moderating effect that corporate response framing has on the relationship between response strategy and E-WoM. Accordingly, our primary outcome variable of interest is consumer E-WoM in response to corporate announcements related to COVID-19 on social media. Our independent variables include categorical variables reflecting response strategies (offensive and defensive) and response framing (rational and emotional), respectively.

We provide an overview of the empirical setting in Figure 2 . In Stage 1, we identified a sample of UK food retailers. To be included in our dataset, the food retailers needed to have more than 250 employees, an active Twitter page and to have made at least one announcement related to COVID-19 on Twitter 4 during the pandemic cri-sis. Using these criteria, a sample of 398 responses to COVID-19 from 50 UK food retailers was identified. Stage 2 involved building the dataset. The independent variables (i.e. response strategy and response framing) were empirically measured using a three-step coding process, while the dependent variables were assessed using social media metrics extracted from Twitter. We then matched corporate announcement data to online consumer comments data to model the relationship between the crisis communication signals and consumer reactions. In the final stage, we conducted a series of analyses to gain insights for crisis communication and service failure recovery and offer practical guidance and outline the implications of crisis communication strategies for food retailers.

Independent variables. The COVID-19 announcements were categorized into emotional framing versus rational framing (Claeys and Cauberghe, 2014) and offensive strategy versus defensive strategy (He, Rui and Whinston, 2018) through rigorous content analysis. Content analysis has been widely adopted in the marketing literature and is an appropriate approach for analysing documents, or statements, that have a clear structure and flow (Kim and Kumar, 2018) , as well as a useful approach for analysing social media posts (Lee, Hosanagar and Nair, 2018; Meire et al., 2019) and other forms of corporate marketing communication signals (Kim and Kumar, 2018) . For example, Kim and Kumar (2018) investigate two specific types of corporate communication signals (i.e. economic and relational) using content analysis to analyse direct corporate marketing communications and examine their impacts on purchase behaviour in a business-tobusiness context. Similarly, Meire et al. (2019) categorize marketer-generated content, extracted from Facebook fan pages, into informational and emotional posts and investigate the types of posts which have had the most influence on consumer sentiments. As marketer-generated content reflects managers' mind-sets and cognitions, it is difficult to utilize a computer-aided text analysis to automatically capture the sharing of meaning, status updates relating to COVID-19, but also to capture consumers' reactions towards corporations (Rao et al., 2020) . complicated information and concepts (Saboo and Grewal, 2013) . Moreover, given the relatively low number of corporate COVID-19 announcements (i.e. 398 announcements) and the absence of ready-to-use dictionaries, we chose manual coding rather than an analytical approach. Indeed, an analytical approach would require training the data to build algorithms and develop a predictive model (Lee, Hosanagar and Nair, 2018) , which is not an effective approach for our study. We provide details of the coding process in Appendix B.

Electronic-word-of-mouth. E-WoM is considered to be an important consumer information source as it consists of consumer opinions, experiences and emotional reactions (Hsu and Lawrence, 2016) . In a public health crisis, E-WoM, generated by individual consumers on social media, has proven to have tremendous impact on corporations, governments and societies, as it is perceived as more trustworthy than traditional media (Park, Boatwright and Avery, 2019) . Thus, using a web crawler, we extracted consumers' 'Reply' tweets which commented on food retailers' COVID-19 crisis response tweets on Twitter, to capture consumer E-WoM.

Control variables. We collected data on three control variables: (1) diversity of response strategy; (2) diversity of response framing; and (3) intensity of response. We detail the measurement of both E-WoM and control variables in Appendix C.

A two-way ANCOVA was employed to analyse the data and test the hypotheses. With regards to control variables, we found that including diversity of response strategy, diversity of response framing and intensity of response measures as covariates in the analyses did not influence the 10 Y. Wang et al. results. The results support the predicted main effect of both corporate crisis response strategy and response framing, as well as the interaction between these factors. As predicted by H1, consumers exposed to tweets by food retailers using a defensive COVID-19 response strategy had higher levels of positive E-WoM, in comparison to consumers exposed to an offensive response strategy (M defensive = 0.1147, M offensive = 0.0106, p < 0.05). Response framing also had a significant effect on E-WoM (supporting H2) as COVID-19 announcements framed in an emotional manner provided more positive E-WoM in comparison to those formulated in a rational manner (M emotional = 0.1917, M rational = 0.0032, p < 0.05). With respect to the hypothesized interactions, we found evidence of a positive interaction between corporate crisis response strategy and response framing (see Table 1 and Figure 3) . Specifically, as posited by H3a, when corporate COVID-19 response announcements posted on social media were developed using a defensive response strategy, the incorporation of a message framed in an emotional manner significantly increased positive E-WoM (M emotional = 0.358, M rational = −0.011, p < 0.05). However, when food retailers used an offensive response strategy, there was no difference between emotional and rational framing in terms of the effect on E-WoM (M emotional = 0.060, M rational = −0.015). Hence, H3b is not supported.

These findings offer novel evidence of the positive impact of utilizing a defensive response strategy when announcing corporate responses to COVID-19 on social media, as consumers had a negative disposition towards offensive announcements during the pandemic crisis. The analysis points out that emotionally framed, corporate crisis responses can provide a significant advantage over rational announcements, with regard to acquiring positive consumer E-WoM. Our results also suggest that corporations should produce crisis responses which extensively utilize a defensive strategy with an emotional focus to improve consumer E-WoM during public health crises.

To check the robustness of our findings, we considered the impact of response timing on the proposed relationships in our model. The results are discussed in Appendix D.

Study 2 builds upon the results of Study 1, by investigating changes in consumers' trust towards different crisis response strategies and framing, formulated by the food retailer brands. We use experimental design procedures to exclude potential factors that might confound the results, such as the established impression of a brand (Aguinis and Bradley, 2014) . In particular, following guidance provided by Aguinis and Bradley (2014) , an experimental vignette methodology (EVM) study is designed and implemented.

Adopting EVM to further examine the hypotheses is appropriate for this study because it is important to exercise control of the interactive effects of two independent variables (i.e. corporate crisis response strategies and response framing) when gathering evidence about causation (Aguinis and Bradley, 2014) . That is, examining whether the interface between corporate crisis response strategies and framing has a causal relationship with changes in consumer trust. In addition, the context of this study focuses on the social media announcements of food retailers and the reactions of consumers in tweets. Adopting an online questionnaire to present the vignettes for simulating social media announcements has strong similarity with the natural setting, as customers normally read tweet messages from either a computer or a mobile phone. Therefore, EVM is a suitable approach for this study (Lohrke, Holloway and Woolley, 2010) . As shown in Appendix E, a three-stage experiment was designed to consider the vignettes of crisis situations and tweet responses associated with a UK food retailer.

The sampling frame of the experiment is UK residents who shop with UK food retailers and are familiar with the use of social media. We developed a set of filter questions to assess social media familiarity, as well as social media usage, to qualify the respondents and ensure that the presented scenario was familiar to them, thereby ensuring internal validity (Aguinis and Bradley, 2014) . Respondents were asked to respond to a yes/no question (i.e. I am familiar with the use of social media platforms for information searching and online communication, such as Facebook, Twitter and Instagram) (Gefen et al., 2003) . Respondents were asked 'how much time, on average, do you spend on social media each day?' with ordinal scale 0 = don't use social media daily; 1 = less than 1 hour; 2 = 1 to 3 hours; 3 = 3 to 5 hours; 4 = 5 to 10 hours; 5 = more than 10 hours. All respondents were familiar with social media and used social media daily, with 86.59% of them indicating that they used social media for at least 1 hour per day. 180 respondents were recruited from the author's social media networks (n = 85, 47.5% of the entire sample) and Amazon's Mechanical Turk (n = 94, 52.5% of the entire sample). We used a t-test to check for significant differences between the various recruitment methods. The results (see Appendix F) indicate that there are no significant differences for (1) 12 Y. Wang et al. demographic background (i.e. gender and age), (2) frequency of daily social media usage and (3) their baseline trust towards the brand across the two different recruitment methods.

The respondents' profiles across eight scenarios are summarized in Table 2 . Chi-square analysis examined the randomization across experimental conditions. The results indicate that the gender ratio [χ 2 (7, N = 179) = 11.9609, p = 0.102] and the respondents' age group [χ 2 (14, N = 179) = 31.1440, p = 0.311] are not significantly different across eight vignettes.

In Study 2, the dependent variable is the recovery of trust from a corporate service failure associated with the COVID-19 crisis. Consumer trust recovery was operationalized as the trustworthiness level at the different stages of the crisis (baseline, during crisis and after seeing tweets). We directly adopted an existing scale: (a) 'I trust this brand'; (b) 'This brand is reliable'; (c) 'This brand is honest with me'; (d) 'This brand is safe', developed by Chaudhuri and Holbrook (2001) . The measurement items were assessed on the five-point Likert scale with 1 = 'strongly disagree' for Stage 1 and Stage 3, with all measurements following the crisis vignettes. The trustworthiness scale of Stage 2 was reverse coded to check respondents' attention for the purpose of manipulation checks (Antonetti and Maklan, 2016) . Across all three experimental stages, the trustworthiness items were highly correlated and reliable. Cronbach's α and means, as well as standard deviation (SD) for trustworthiness at all stages and vignettes are presented in Table 3 .

The experiment was designed to scrutinize how corporate social media announcements influence consumer trust recovery during a public health crisis and subsequently, after corporate tweet intervention. The two different crisis scenario groups did not significantly differ regarding the trustworthiness scores at all three stages. The results of three paired sample t-tests comparing two scenario groups are presented in Table 3 .

Consumer trust declined significantly between Stage 1 (baseline) and Stage 2 (during the crisis) scenarios of product shortage, t(91) = 10.86, p < 0.001 and social distancing, t(86) = 11.07, p < 0.001. Following the design of Van Norel et al. (2014) in their study of damaged corporate reputation, we tested the overall efficiency of using tweets to communicate with customers and recover trust that was damaged during the COVID-19 crisis. On average, customer trust recovers significantly and the use of appropriate tweets can build trust to levels which are higher than pre-crisis levels, t(178) = 16.13, p < 0.001. To test the hypothesis in Study 2, we conducted a 2 × 2 (response strategy × response framing) analysis of variance with crisis scenario (product shortage and social distancing) as control variable. First, we expected the use of defensive and offensive crisis response strategies to have significantly different impacts on trust recovery. However, our analysis rejected H4, as a defensive crisis response strategy did not result in greater consumer trust recovery compared to an offensive response strategy, F(1, 178) = 0.00, n.s., η 2 = 0.00. Second, the main effect for the response framing group was significant, F(1, 178) = 11.37, p < 0.001. Respondents exposed to tweet announcements with an emotionally framed response have significantly greater consumer trust recovery than those exposed to a rationally framed corporate crisis response. More specifically, the tweets framed in an emotional manner enabled the trustworthiness scores to recover from 2.31 to 4.09, while those framed in a rational manner improved the trustworthiness score from 2.54 to 3.69. Interestingly, a rational crisis framing response did not result in customer trust recovering to a pre-crisis level, while an emotional crisis framing response led to trust levels that were stronger than pre-crisis levels. Therefore, H5 is supported.

Finally, as implied by H6a, the response framing × response strategy interaction was significant, F(1, 178) = 18.60, p < 0.001, η 2 = 0.07. According to simple effect analysis, we find that a corporate crisis, defensive response strategy communication announcement on twitter, which is framed in an emotional way, will result in higher levels of trust recovery than an announcement framed in a rational manner. There is a significant effect of response framing for a defensive strategy, F(1, 174) = 24.25, p < 0.001. On average, the defensive response strategy with rational framing increased trustworthiness scores from 2.75 to 3.58 (i.e. trust recovery = 0.83), while the trust recovery has a greater change (i.e. average consumer trust scores changed from 2.21 to 4.26) when the defensive strategy is framed in an emotional manner (i.e. trust recovery = 2.05). However, as shown in Figure 4 , emotional framing and rational framing did not result in significant differences in trust recovery when an offensive strategy was used in corporate tweet communications F(1, 174) = 0.09, p = 0.76. The trust recovery made by the offensive strategy with emotional framing is 1.41 (i.e. average consumer trust scores changed from 2.45 to 3.86) and with rational framing, 1.48 (i.e. average consumer trust scores changed from 2.32 to 3.80). Thus, H6b is rejected.

Effective crisis communication responses are crucial for corporations seeking to successfully recover from service failures induced by a public health crisis. Our study sets out to explore how corporate crisis response strategy and response 14 Y. Wang et al. Figure 4 . Two-way interaction effect of response strategy and response framing on trust recovery [Colour figure can be viewed at wileyonlinelibrary.com] framing can be deployed to shape positive consumer E-WoM and rebuild consumer trust. Drawing upon the results of two studies, we answer three research questions. Theoretically, we provide new insights into corporate social media announcements in a public health crisis by analysing how response strategy (offensive vs. defensive) and response framing (emotional vs. rational) drive positive consumer E-WoM and trust recovery, and by exploring the interaction effects between corporate crisis response strategy and framing. Methodologically, this paper makes a timely response to the recent calls for future research, focusing on service failure and recovery in a public health context which analyses rich social media collected from multiple, rather than single, sources (Grégoire and Mattila, 2020; Khamitov, Grégoire and Suri, 2020) . We also provide useful practical insights that will enable corporations to strategize crisis communication efforts in order to obtain positive consumer E-WoM and recover lost trust and highlight policy implications.

The theoretical implications of this study to signalling theory and the service failure and public health crisis literature are threefold. First, although many studies adopting signalling theory have investigated the characteristics of signals (e.g. volume, timing, strength, clarity and visibility) (Hsu and Lawrence, 2016; Ni, Flynn and Jacobs, 2016) , less attention has been paid to explore how to formulate signals effectively. To address this, our study builds on the extant service failure (e.g. Claeys and Cauberghe, 2014; Kharouf et al., 2020) and public health crisis (Bunker, 2020; Novelli et al., 2018) literature by highlighting the importance of strategizing and framing corporate signals in response to a global pandemic crisis. We further enrich the current literature by providing new evidence to show that a defensive response strategy, framed in an emotional manner, contributes to the formation of positive consumer E-WoM and the recovery of consumer trust during a public health crisis. This finding contrasts with prior research on service failure, which concluded that rational framing would result in more positive, post-crisis consumer attitudes towards a corporation (Claeys and Cauberghe, 2014) .

Second, understanding how receivers (i.e. consumers) translate signals into perceived meaning is considered to be one of the most important elements of signalling theory (Connelly et al., 2011) . In contrast to previous studies which have examined consumers' interpretation of a service failure event by using the measurement scales of perceived service quality (Gijsenberg, Van Heerde and Verhoef, 2015) , attitude towards service (Liu, Pennington-Gray and Krieger, 2016) and satisfaction (Kharouf et al., 2020) , we introduce E-WoM as a reflection of a corporation's crisis communication effectiveness in a public health crisis context. We do this by building upon research that has used social media data to investigate consumer E-WoM (e.g. Aleti et al., 2019; Wang et al., 2020) . Our measurement of E-WoM allows researchers to better quantify consumer reactions and deeply capture consumers' positive, or negative, interpretation of corporate crisis communication responses during the pandemic. In doing so, we answer recent calls for further research (Grégoire and Mat-tila, 2020) , which suggest using social media data to capture consumer reactions during a service failure.

Third, the signalling environment is one of the core components in signalling theory. Research on public health crisis communication has explored the effects of signalling in organizational and institutional environments. Yet, less attention has been paid to corporate crisis responses or the social media environment when addressing public health crises. By focusing on a social media environment, where signalling occurs in a noisier and more complex context than previous research, we provide novel insights into the crisis communication literature by explaining the varied effects of signals on consumer reactions on social media.

This study offers actionable corporate crisis communication strategies for practitioners and policymakers. More specifically, our findings offer support for campaigns which are similar to this and suggest that corporations should consider a shift in emphasis, away from traditional offensive and rational crisis communication responses, towards more defensive and emotional ones. By doing so, corporations will have a higher likelihood of service failure recovery, acquiring positive consumer E-WoM and rebuilding customer trust during and after a crisis. Indeed, the retailing sector has experienced the phenomenon of consumers stockpiling and bulk-buying during the early stages of the COVID-19 outbreak and consumers continue to be concerned for their safety. Crisis response campaigns may be employed to reduce consumers' anxiety and provide emotional support related to the crisis. A recent example is Cottonelle's 'ShareASquare' social media campaign. As a toilet paper producer, their campaign, in partnership with charities, delivers emotional messages on Twitter to encourage people to be generous during this frustrating time. Cottonelle donates US$1 when consumers use the hashtag #ShareASquare to share their announcements.

Second, COVID-19 has caused untold consumer suffering including financial difficulties, food shortages and long queues while shopping. With a lack of clear guidance on how to respond to customers, corporations could potentially lose consumer trust, even though they might be considered 'victims' (Coombs, 2007) , since the crisis was not caused by them. Our findings provide insights into the effects of different corporate crisis responses, and therefore offer direct guidance for corporations seeking to formulate appropriate communication strategies. To obtain positive E-WoM and restore consumer trust during the post-crisis period, corporations should effectively utilize a defensive approach when developing response strategies. Moreover, social media environments provide flexible opportunities for corporations seeking to adjust and implement crisis response strategies. Our findings offer guidance to practitioners seeking to develop social media crisis communication strategies in the food retailing sector and confirm that timely corporate communication can generate attention and influence consumer reactions.

Third, our research also provides potential implications for UK policy-makers who have been criticized by political commentators and others for presenting mixed messages relating to how to stay safe during the COVID-19 pandemic (Tolhurst, 2020). As some members of society have not been following social distancing and other guidelines, a better understanding of the types of interventions and public health messages that would lead to an increase in compliance is essential for policy-makers (Norman, Wilding and Conner, 2020) . Our findings are closely aligned with the latest UK Parliament's COVID-19 'Areas of interest-communication strategy for public health messages' (Bermingham, 2020) . Specifically, the findings obtained from this study can contribute to the discussion of the key research question: 'How do different approaches to communicating uncertainty affect people's likelihood to follow guidance?' (Bermingham, 2020) . Based on the empirical evidence we obtained, we suggest that policy-makers should consider the potential effect of response framing when developing public health communication strategies shared in social media, aimed at engaging citizens and encouraging them to follow measures and guidance (e.g. social distancing or following self-isolation rules if they have potentially been exposed to the virus or abroad). We would encourage government authorities to carry out A/B testing and evaluate people's perceptions and attitudes towards public messages before releasing announcements on social media.

Y. Wang et al. Limitations and directions for future research Several opportunities exist for future research that will help overcome the limitations associated with this study. Firstly, the single country and industry context of this study may mean that our findings are not generalizable to other sectors or regions. It would therefore be valuable for future research to validate the current findings by collecting and analysing social media data from different business sectors in a broader range of countries or regions.

Second, although our selected outcome variables are valuable for achieving our research objective, they do not fully reflect the consequences of corporate crisis communication. Future research could explore additional potential consequences of corporate communication responses to public health crises, such as how crisis communication strategies influence brand engagement or, in the longer term, have an impact on corporate financial performance (e.g. cash flow and actual sales).

Third, as this study merely considers response framing as the moderating role, research might further examine the contingent role of signalling in the signalling process. We hope that this study sparks future research on exploring various characteristics of signals-such as consistency or enabling mechanisms of signalling process (e.g. consumer engagement and brand involvement). These characteristics and mechanisms might generate different influences on consumer behavioural outcomes when used by different senders.

Finally, the use of Twitter data alone may create a limitation, as Twitter may provide different insights from other platforms such as Facebook, or online reviews, and does not capture all corporate communication signals. Although some researchers have claimed that brand-related sentiment in user-generated content does not differ across social networking sites (Smith, Fischer and Yongjian, 2012) , future studies which investigate the effect of corporate responses to public health crises on consumer sentiment across different social media platforms would be beneficial.

Best practice recommendations for designing and implementing experimental vignette methodology studies

Tweeting with the stars: automated text analysis of the effect of celebrity social media communications on consumer word of mouth

Social identification and corporate irresponsibility: a model of stakeholder punitive intentions

Online business reporting: a signaling theory perspective

Coronavirus: Boris Johnson says UK is past the peak of outbreak

COVID-19 areas of research interest

The response to Covid-19 in England: political accountability and loss of trust

Who do you trust? The digital destruction of shared situational awareness and the COVID-19 infodemic

The chain of effects from brand trust and brand affect to brand performance: the role of brand loyalty

A crisis of epidemic proportions: what communication lessons can practitioners learn from the Singapore SARS crisis?

What makes crisis response strategies work? The impact of crisis involvement and message framing

Implications of stealing thunder for the impact of expressing emotions in organizational crisis communication

Restoring reputations in times of crisis: an experimental study of the situational crisis communication theory and the moderating effects of locus of control

Signaling theory: a review and assessment

Consumers and the food system under Covid-19

Protecting organization reputations during a crisis: the development and application of situational crisis communication theory

An exploratory study of stakeholder emotions: affect and crises

Tailoring service recovery messages to consumers' affective states

Toward a cognitive view of signalling theory: individual attention and signal set interpretation

E-WoM and accommodation: an analysis of the factors that influence travelers' adoption of information from online reviews

Defensive marketing strategy by customer complaint management: a theoretical analysis'

A bibliometric investigation of service failure literature and a research agenda

Affect versus cognition in the chain from perceived quality to customer loyalty: the roles of product beliefs and experience

Overcoming institutional voids: a reputation-based view of long-run survival

Trust and TAM in online shopping: An integrated model

Losses loom longer than gains: modeling the impact of service crises on perceived service quality over time

Why the global recession could last a long time

Service failure and recovery at the crossroads: recommendations to revitalize the field and its influence

Defensive marketing strategies

The impact of Covid-19 pandemic on corporate social responsibility and marketing philosophy

Social media strategies in product-harm crises

Service failure and recovery: the impact of relationship factors on customer satisfaction

The role of social media and brand equity during a product recall crisis: a shareholder value perspective

Responding to a health crisis on Facebook: the effects of response timing and message appeal

Defensive strategy framework in global markets

A systematic review of brand transgression, service failure recovery and product-harm crisis: integration and guiding insights

A signaling theory approach to relationship recovery

The relative influence of economic and relational direct marketing communications on buying behavior in business-to-business markets

Removing the shadow of suspicion: the effects of apology versus denial for repairing competence-versus integrity-based trust violations

Growth opportunities for brands during the COVID-19 crisis

Aldi to create 4,000 more UK jobs as sales surge in Covid crisis', The Guardian

Advertising content and consumer engagement on social media: evidence from Facebook

Embedding trust into COVID-19 recovery: four dimensions of stakeholder trust

Tourism crisis management: can the extended parallel process model be used to understand crisis responses in the cruise industry?

Conjoint analysis in entrepreneurship research: a review and research agenda

Neglected outcomes of customer satisfaction

Crisis communication and public perception of COVID-19 risk in the era of social media

Bioterrorism and local agency preparedness: results from an experimental study in risk communication

Offensive and defensive marketing in spatial competition

The state of social media research: where are we now, where we were and what it means for public relations

The role of marketer-generated content in customer engagement marketing

The commitment-trust theory of relationship marketing

Corporate reputation: do board characteristics matter?

The effect of a toy industry product recall announcement on shareholder wealth

Reasoned action approach and compliance with recommended behaviours to prevent the transmission of the SARS-CoV-2 virus in the UK

No Ebola… still doomed" -the Ebola-induced tourism crisis

Mitigating loneliness with companion robots in the COVID-19 pandemic and beyond: an integrative framework and research agenda

Cross-national differences in consumer response to the framing of advertising messages: an exploratory comparison from Central Europe

A case study of best practice -Visit Scotland's prepared response to an influenza pandemic

Information channel preference in health crisis: exploring the roles of perceived risk, preparedness, knowledge, and intent to follow directives

The nature and role of feedback text comments in online marketplaces: implications for trust building, price premiums, and seller differentiation

Retweets of officials' alarming vs reassuring messages during the COVID-19 pandemic: implications for crisis management

Tourism disaster planning and management: from response and recovery to reduction and readiness', Current Issues in Tourism

Defensive marketing

Stock market reactions to customer and competitor orientations: the case of initial public offerings

A typology of consumers' emotional response styles during service recovery encounters

We are not the same as you and I: causal effects of minor language variations on consumers' attitudes toward brands

Business of business is more than business: managing during the Covid crisis', Industrial Marketing Management

How does brandrelated user-generated content differ across YouTube, Facebook, and Twitter?

Market snapshot: a report of the UK's retail industry

Go out, stay at home, get a test, don't get a test: the full story of the government's mixed coronavirus messaging

Insight from the horsemeat scandal: exploring the consumers' opinion of tweets toward Tesco', Industrial Management & Data Systems

Emotional crisis communication

Damaged corporate reputation: can celebrity tweets repair it?

Negative online word-of-mouth: behavioral indicator or emotional release?

Communicating tourism crises through destination websites

Examining the antecedents and consequences of corporate reputation: a customer perspective

Relationship between online retailers' reputation and product returns

Unpacking the impact of social media analytics on customer satisfaction: do external stakeholder characteristics matter?

Why marketers don't market: rethinking offensive and defensive archetypes

Resolving a dilemma of signaling bankrupt-firm emergence: a dynamic integrative view

Moving forward: the effectiveness of online apologies framed with hope on negative behavioural intentions in crises

The impact of expressing mixed valence emotions in organizational crisis communication on consumer's negative wordof-mouth intention

How to repair customer trust after negative publicity: the roles of competence, integrity, benevolence, and forgiveness

His research interests include big data analytics and social media marketing. His research has been published in journals such as the British Journal of Management

His research interests include managerial risk perception, supply chain risk management, quality management and social media analytics. Thanks to various cross-disciplinary research collaborations, Minhao has published papers in the fields of supply chain management, environmental management and big data

She is currently a university teacher in Information and Operation Management in the Management School at the University of Sheffield. Her research interests include management of information systems, and organizational information and knowledge management. Her work has been published in Technological Forecasting and Social Change and Industrial Management and Data Systems

Fraser McLeay's current research interests involve social media and communities of consumption, hedonic versus sustainable consumption and the role that digital marketing can play in driving consumers to adopt innovative nascent technologies. His research has been published in the Journal of Service Research, Journals in Marketing Management, Tourism Management

Additional supporting information may be found online in the Supporting Information section at the end of the article.Appendix