Do memorable restaurant experiences affect eWOM? The moderating effect of consumers' behavioural engagement on social networking sites

Purpose – Many restaurants offer high-quality service to their customers, hoping to provide memorable experiences that influence their loyalty and electronic word of mouth (eWOM). However, consumers ’ memorableexperiencesdonotalwaysimplypositiveeWOM.Thisstudyaimsto(1)verifythedirectimpactsoftheperceivedqualitybyconsumersofcasualdiningrestaurantsonpositiveemotions,negativeemotionsand memorableexperiences;(2)investigatetheimpactsofmemorableexperiencesonthepropensitytoloyaltyandeWOM;(3)testthemoderatingeffectofconsumerbehaviouralengagementonsocialnetworkingsites(CBE-SNS)ontherelationshipbetweenmemorableexperiencesandeWOM. Design/methodology/approach – This survey included 475 university students in Brazil. Participants answered an electronic form about their experiences in casual dining restaurants. Structural equation modelling tested the hypothetical model based on the stimulus-organism-response (S-O-R) theory (Mehrabian and Russell, 1974). Findings – The quality perceived by restaurant consumers (stimulus) positively impacts their memorable experiences and positive emotions and negatively affects their negative emotions (organism). Memorable experiencespositivelyimpactthepropensitytoloyalty(response).TheCBE-SNSmoderatestheintensityofthe relationshipbetweenmemorableexperiences(organism)andeWOM(response). Originality/value – This study is the first that demonstrates the relationships between perceived quality, positive and negative emotions, memorable experiences, the propensity to loyalty and CBE-SNS and e-WOM in restaurants.Casualdining restaurantsmust offer their customers services withhigh perceived quality, positively impactingtheiremotionsandtheirmemorableexperiences.Finally,restaurantsmustcreatestrategiesandactionstoincreasetheCBE-SNStoencouragethemtosharetheirmemorableexperiencesthrougheWOM.

Do memorable restaurant experiences affect eWOM? The moderating effect of consumers' behavioural engagement on social networking sites 1. Introduction A memorable experience is an event constructed from real individuals' experiences stored in the memory and recalled later. Further, a memorable experience is one that a consumer considers unique and remains in the memory over time (Servidio and Ruffolo, 2016). Providing memorable experiences to consumers is a top priority in many industries (Oh et al., 2007). It is particularly crucial in services such as tourism (Rasoolimanesh et al., 2021;Sthapit and Coudounaris, 2018;Servidio and Ruffolo, 2016), hospitality (Ali et al., 2014) and food service (Cao et al., 2019;Stone et al., 2018). For this reason, several authors have studied the various aspects of memorable experiences, their antecedents and consequences (Bastiaansen et al., 2019;Fernandes et al., 2015). The quality perceived by consumers has been considered an antecedent of emotions, memorable experiences and attitudes and behavioural intentions (Ahmed et al., 2022;Rajput and Gahfoor, 2020;Cao et al., 2019;Stone et al., 2018). Among the attitudes and behavioural intentions most studied in consumer behaviour is the propensity to loyalty (Horng and Hsu, 2021;Souki et al., 2020;Ali et al., 2014;Oliver, 1999;Zeithaml et al., 1996), word of mouth (WOM) communication, and electronic word of mouth (eWOM) communication (Lin et al., 2022;Chen et al., 2022;Line et al., 2020;Tussyadiah et al., 2018;Hwang, 2018;Hennig-Thurau et al., 2004;Ratchford et al., 2001;Stauss, 1997). Lin et al. (2022) tested the impacts of consumers' perception of quality concerning their restaurant experiences on arousal and memorisation. They also assessed if arousal and memorability affect WOM and eWOM. This study reveals that consumer memorability impact WOM more than eWOM. Hence, it suggests that consumer behavioural engagement in social networking sites (CBE-SNS) can affect the strength of the relationship between memorable experiences and eWOM. Souki et al. (2020) proposed and tested a comprehensive model to assess the impacts of nine dimensions of perceived quality by consumers of a la carte restaurants on their attitudes and behavioural intentions. These authors argue that a la carte restaurants offer their consumers food and drinks at individual prices for each dish on the menu. So, consumers can choose dishes and drinks from the menu separately rather than ordering a pre-set combination of items at a fixed price. Lim et al. (2022) highlight that, among the a la carte restaurants, some are pretty sophisticated (fine dining) and others are more informal (casual dining). Shen et al. (2021) affirm that, like fine dining restaurants, casual dining restaurants offer full service, with waiters serving customers at tables and food and drinks on an individually priced list on the menus. However, casual dining restaurants charge more affordable prices, allowing frequent visits from different audiences (e.g. families, friends and students). The present study contemplates the casual dining restaurants consumers' behaviour.
Considering the above, the following guiding questions for this study are: (1) Does the quality perceived by casual dining restaurant consumers' impact their positive and negative emotions and memorable experiences?
(2) Customers with memorable experiences in casual restaurants tend to amplify their propensity to loyalty?
(3) Consumers who have memorable experiences at casual restaurants tend to do eWOM?
(4) Does CBE-SNS influence the strength or direction of the relationship between their memorable casual dining experiences and eWOM?
To answer this study's guiding questions, the authors recurred to the stimulus-organismresponse (S-O-R) theory, proposed by Mehrabian and Russell (1974). This theory assesses the sequential relationships between environmental stimuli, emotional and cognitive states, and TQM consumers' responses and behaviours.  and Brewer and Sebby (2021) highlight that the S-O-R theory advocates that physical or social stimuli directly affect people's emotional and cognitive states, influencing their subsequent behaviours. The S-O-R theory has been frequently used in research on consumers' experiences in food services (Lim et al., 2022;Brewer and Sebby, 2021;Oh and Kim, 2021). However, the S-O-R theory has not been used in previous studies to describe the relationships between (1) environmental stimuliquality perceived by consumers of casual dining restaurants; (2) organismcognitive (memorable experiences) and emotional states (positive and negative emotions), and; (3) behavioural responses -eWOM and propensity to loyalty. Furthermore, no previous study has included the casual dining restaurants CBE-SNS as a moderator of the relationship between memorable experiences (organism) and eWOM (response). Therefore, the present research focuses on filling this gap in the literature.
This study aims to (1) verify the direct impacts of the perceived quality by consumers of casual dining restaurants (stimulus) on positive emotions, negative emotions and memorable experiences (organism); (2) investigate the impacts of memorable experiences (organism) on the propensity to loyalty and eWOM (response); (3) test the moderating effect of CBE-SNS on the relationship between memorable experiences (organism) and eWOM (response).
This study contributes academically by using the S-O-R theory to demonstrate the direct impacts of the perceived quality by consumers of casual dining restaurants (stimulus) on memorable experiences and positive and negative emotions (organism) and the subsequent effects on eWOM and the propensity to loyalty (responses).
Another theoretical contribution of the present study is to demonstrate that the CBE-SNS is an independent construct that moderates the relationship between consumers' memorable experiences in casual restaurants (organism) and eWOM (response). Therefore, adopting digital technologies affect the relationship between consumers' memorable experiences and their eWOM. This result suggests that future studies should consider variables that can moderate the relationship between stimulus and organism and between organism and responses.
This study also contributes managerially by demonstrating that casual dining restaurants must offer their customers tangible and intangible attributes that influence a high perception of quality, positively impacting their emotions, memorable experiences and propensity to loyalty. However, it reveals that customers who favourably perceive the quality of their restaurant experiences do not always show positive eWOM. Hence, restaurant managers should consider the CBE-SNS to define different strategies for each public profile and positively impact eWOM.

Theoretical background and research hypotheses
The present study used the S-O-R theory to demonstrate sequential causal relationships between the constructs of the hypothetical model ( Figure 1). The perceived quality factors (stimulus) used in this study are infrastructure, food quality, accessibility and convenience, customer orientation, service quality, atmosphere, social endorsement, reputation and status (Souki et al., 2020). Also, this study tests whether these perceived quality factors impact consumers' positive and negative emotions and memorable experiences (organism). The model also verifies if consumers' memorable experiences are an antecedent of their propensity to loyalty and eWOM (response). Finally, this study tests the moderating effect of CBE-SNS on the relationship between consumers' memorable experiences (organism) and eWOM (response).
2.1 Perceived quality (stimulus) and its relationship to consumers' memorable experience (organism) Perceived quality refers to consumers' perception of the performance of products or services in attributes capable of satisfying their needs and expectations compared to competitors Memorable restaurant experiences affect eWOM (Souki et al., 2020;Zeithaml, 1988). Companies in tourism, hospitality and restaurants must continually monitor the quality perceived by consumers, as it is a critical component to keep them competitive (Ahmed et al., 2022;Rajput and Gahfoor, 2020).
However, in the context of restaurants, perceived quality needs to go beyond the apparent aspects such as infrastructure, accessibility or good food (Tripathi and Dave, 2016) and need to consider intangible elements such as service quality, restaurant atmosphere, social endorsement, status and consumer orientation (Souki et al., 2020). Pine and Gilmore (1998) suggest that organisations should intentionally create memorable events to engage customers. Thus, restaurant managers and academicians seek to identify the tangible and intangible perceived quality factors that can enhance consumers' memorable experiences (Cao et al., 2019;Stone et al., 2018).
Restaurants must create environmental conditions to provide consumers with unique, positive, and memorable experiences (Fernandes et al., 2015). T€ urker et al. (2019) point out that the servicescape attributes are crucial to creating a unique atmosphere to make gastronomic experiences unforgettable, increasing consumer satisfaction and loyalty. Fernandes et al. (2015) argue that sensory stimuli are crucial tools for marketing as they generate unique and memorable consumption experiences. Hult en et al. (2009) corroborate this statement, pointing out that the five human senses (sight, sound, smell, taste and touch) give invaluable information about different things and are indispensable to human existence. Through the senses, the human being captures the stimuli from the environment (e.g. sounds, images, temperatures, textures, smells and tastes), generating emotions, thoughts, comprehension, learning and memories of their experiences (Stone et al., 2018). Pine and Gilmore (1998) and Oh et al. (2007) advocate that the ideal is to engage all five senses because the more an experience involves, the more exciting and memorable it tends to be. Lin et al. (2022) demonstrated that TQM the perceived quality by consumers in restaurants directly and positively impacts their memorable experiences. Hence, the following hypothesis is: H1. Perceived quality directly and positively impacts the consumers' memorable experience in casual dining restaurants.
2.2 Perceived quality (stimulus) and its impact on positive and negative emotions (organism) According to Bagozzi et al. (1999), emotions are a mental state of readiness derived from cognitive evaluations of events or thoughts accompanied by physiological processes typically expressed physically by facial expressions, postures and gestures, among others. Bastiaansen et al. (2019) mention that emotions are responses of biological origin and constitute the main driving force of human behaviour. These authors also point out that emotions have three levels related to the way people feel (subjective experience), how they react to certain stimuli (expressive behaviour) and the physical changes that occur in response to the mentioned stimuli (physiological body changes). Oliveira et al. (2022) argue that consumers' emotional responses to perceived quality during their consumption experiences can be positive or negative. Oh and Kim (2021) developed an investigation into the role of emotions in online fine-dining restaurant reviews. Based on Herzberg's two-factor theory, some aspects of consumer experiences generated only positive emotions, while others generated negative emotions. Therefore, it is necessary to include separate items in the surveys to measure positive and negative emotions (Jung et al., 2021).
However, Song and Kim (2021) highlight that research on consumer behaviour in tourism, hospitality and restaurants has focused mainly on positive emotions. These authors investigated the cause-effect relationship between harmful food incidents and tourists' negative emotions. They cite measurement items typically used in food service to measure negative emotions, such as anger, distress, disgust, fear, shame, hostility, irritation, nervousness, fear, disappointment, regret, worry and anxiety. Leung and Wen (2021) examine the role of consumer emotions in digital food ordering experiences. To this end, they included separate measurement items for positive emotions (e.g. happy, relaxed, comfortable and passionate, among others) and negative emotions (e.g. frustrated, annoyed, disappointed and angry, among others). Souki et al. (2020) also measured consumers' positive and negative feelings in their study of a la carte restaurants. Among the negative emotions, these authors used the following indicators: anger, irritation, frustration, sadness and discontent. On the other hand, they used the following indicators to assess positive emotions: happy, excited, calm, optimistic and enthusiastic. Lin et al. (2022), Shahzadi et al. (2018) and Marinkovic et al. (2014) claim that the quality perceived by customers, concerning their experiences in restaurants influences their emotions. Thus, the following hypotheses are: H2. Perceived quality positively and directly impacts casual dining restaurant customers' positive emotions.
H3. Perceived quality negatively and directly impacts casual dining restaurant customers' negative emotions.
2.3 Positive and negative emotions and their impact on consumers' memorable experiences (organism) Oliveira et al. (2022) and Oliver (1999) (Servidio and Ruffolo, 2016;Oh et al., 2007). For this reason, these authors consider emotions a relevant experiential variable to measure. Exploring emotions and the human senses is pivotal to generating unforgettable restaurant experiences. Lin et al. (2022) and Horng and Hsu (2021) claim that emotions impact consumers' memorable restaurant experiences. Based on the above, the following hypotheses are proposed: H4. Positive emotions directly and positively impact consumers' memorable experiences in casual dining restaurants.
H5. Negative emotions directly and negatively impact consumers' memorable experiences in casual dining restaurants.
2.4 Memorable experiences (organism) and the propensity to loyalty (response) Oliver (1999, p. 34) defines loyalty as "a deeply held commitment to rebuy or repatronize a preferred product/service consistently in the future, thereby causing repetitive same-brand or same brand-set purchasing, despite situational influences and marketing efforts having the potential to cause switching behavior". Souki et al. (2020) mention that propensity to loyalty is a consumer's behavioural intention to maintain a relationship with certain companies or brands, buying their products and services over time, even when receiving offers from competitors. Blackwell et al. (2011) argue that during the buying decision process, consumers seek information from both internal (memory) and external sources (media and influence groups). Memory is typically the first source of information used by consumers for their buying decisions. Sthapit and Coudounaris (2018) affirm that memory is the most critical information source for an individual when making a revisit decision and spreading WOM information. Triantafillidou and Siomkos (2014) point out that unique and memorable dining experiences can create value for companies. Consumers' memories of their experiences influence their decision-making processes, behavioural intentions and WOM communications. Previous studies demonstrate the direct and positive relationship between memorable experiences and consumer loyalty in tourism (Rasoolimanesh et al., 2021;Servidio and Ruffolo, 2016), hotels (Ali et al., 2014) and restaurants (Horng and Hsu, 2021). Therefore, the following hypothesis is: H6. The consumers' memorable experiences in casual dining restaurants directly and positively impact their propensity to loyalty.

Memorable experiences (organism) and eWOM (response)
Communication between consumers through the internet has been an object of interest to managers and academics since the mid-1990s. Stauss (1997) states that, with the exponential growth of the internet, consumers have had opportunities to communicate with others about their consumer experiences on a global level. Ratchford et al. (2001) argue that consumers can obtain information about goods and services from friends, acquaintances, colleagues and even unknown people. Thus, consumers seek and provide information about products, services and brands through the internet (Stauss, 1997). However, the most widespread concept of eletronic WOM communication (eWOM) in academia is the one proposed by Hennig-Thurau et al. (2004, p. 39). These authors define eWOM as "any positive or negative statement made by potential, actual or former consumers about a product or company, which is made available to a multitude of people and institutions via the internet". Souki et al. (2022) argue that consumers have become more engaged and proactive in creating and exchanging content through SNS. Thus, consumers can obtain information from other people or companies, evaluate and comment on their consumption experiences, and TQM share content on SNS through eWOM (Chen et al., 2022;Oliveira and Casais, 2019;Serra-Cantallops et al., 2018). Hence, new technologies allow restaurant consumers to obtain information and share their experiences, knowledge and positive or negative opinions through eWOM, anywhere and anytime (Sann et al., 2020;Sahelices-Pinto et al., 2018;Shamhuyenhanzva et al., 2016). Hwang (2018) argues that consumers use smartphones to share their gastronomic experiences through posts with food and restaurant infrastructure images and write about their experiences. Tussyadiah et al. (2018), Hwang (2018) and Shamhuyenhanzva et al. (2016), in their studies on restaurants, claim that opinions, experiences and content shared on the SNS by other consumers, opinion makers, acquaintances and even unknown people tend to generate more credibility and trust than information published by companies. Therefore, other users' opinions and recommendations are crucial in food service (Sann et al., 2020;Oliveira and Casais, 2019). Finally, Chen et al. (2022) and Line et al. (2020) highlight that monitoring the eWOM is pivotal for restaurant managers. Rasoolimanesh et al. (2021) found that heritage tourists' memorable experiences positively impact their eWOM intentions. Lin et al. (2022) demonstrated that restaurant consumers' memorable experiences directly and positively impact eWOM. Hence, the following hypothesis is formulated: H7. Consumers' memorable experiences in casual dining restaurants directly and positively impact their eWOM.

Moderating effect of consumers' behavioural engagement on social networking sites on the relationship between memorable experiences (organism) and eWOM (response)
According to Dessart (2017), consumer engagement in the SNS encompasses cognitive, affective and behavioural dimensions. This author highlights that behavioural engagement refers to the most active manifestations of its concept. Correia et al. (2018) state that the CBE-SNS includes following and liking contacts' posts, commenting on posts, sharing content published by others and creating and publishing content, including photos, videos and texts on social networks. Bailey et al. (2021) argue that CBE-SNS can be passive or active (creative). More passive engagement involves reading comments and viewing photos, videos and content other users have created.
On the other hand, active or creative engagement includes writing reviews, posting blogs, making recommendations to other users, and posting videos, audio or images on SNS. Gvili and Levy (2018) argue that consumer engagement in SNS depends on their degree of connection and involvement with digital platforms. Dolan et al. (2016) classify the interaction of consumers in the SNS based on the following criteria: (1) intensity (low-passive or highactive); and (2) valence (ranging from negative to positive). Dessart (2017) states that the CBE-SNS provides a more practical and aligned perspective with the metrics used by managers to measure the performance of companies on SNS. Kanje et al. (2020) suggest the following profiles according to the CBE-SNS: (1) Opinion makersconsumers who participate more actively in online communications, seeking information, sharing their experiences and influencing others on SNS; (2) Opinion seekersconsumers who tend to seek only information and advice, without sharing their experiences on the SNS; (3) Pass-through behaviour: consumers who only share content generated by third parties; (4) Observer: consumers who have low behavioural engagement on the SNS, not seeking or sharing their experiences, but only observing the content shared by their contacts on social networks. Kang and Namkung (2016) state that new information and communication technologies have substantially increased the number of people engaged in social networks, amplifying the potential impacts of eWOM in the restaurant sector. However, consumers have different Memorable restaurant experiences affect eWOM levels of behavioural engagement in the SNS. Thus, it is crucial to understand the conditions that favour the expansion of this engagement and its impacts on eWOM (Rasoolimanesh et al., 2021;Tussyadiah et al., 2018). However, no previous study demonstrated how engagement levels (low or high) affect the direction and/or strength of the relationship between casual restaurant consumers' memorable experiences (organism) and eWOM (behavioural responses). Therefore, the following hypothesis is: H8. Consumers' behavioural engagement in SNS moderates the relationship between their memorable experiences in casual dining restaurants and their eWOM.

Methodology
The present research is quantitative and descriptive, with cross-sectional data collection. Its hypothetical model considered the quality factors perceived by consumers of a la carte restaurants proposed by Souki et al. (2020). This model also included the eWOM constructs (Line et al., 2020;Serra-Cantallops et al., 2018) and consumers' behavioural engagement in SNS (Correia et al., 2018;Dolan et al., 2016). This study's constructs, classified according to the S-O-R Theory (Mehrabian and Russell, 1974) and their respective measurement items, came from previous studies and were evaluated and validated by three marketing and consumer behaviour specialists (Table 1). Google Forms ® was the platform of electronic forms used to collect this survey's data. To test the hypothetical model, the authors developed an electronic questionnaire with 79 items, including the constructs of perceived quality, positive emotions, negative emotions, memorable experiences, eWOM and CBE-SNS. This study also included questions about the socio-demographic profile of respondents.
This research uses a five-point Likert-type agreement or disagreement scale, where one (1) corresponds to "totally disagree" and five (5) represents "totally agree", as recommended by Dedeoglu et al. (2018) and Malhotra et al. (2017). In addition, it includes a sixth answer option ("does not know/does not apply" -DK/NR). Antonialli et al. (2017) state that the Likert-type scales are the most used form in the literature in applied human and social sciences to measure attitudes. Malhotra et al. (2017) point out that Likert-type scales are recommended when participants must indicate their degree of agreement or disagreement with a series of statements about specific stimulus objects. These authors also point out that the five-point Likert scale has several advantages, such as being easy to build and apply and favouring immediate understanding by the respondents. Dedeoglu et al. (2018) corroborate defending the use of five-point Likert scales due to their ease of handling and providing a more straightforward appearance for respondents. Malhotra et al. (2017) consider that five-point Likert scales are suitable for collecting data from online surveys, in kiosks, mobile devices, by mail, by phone or in personal interviews. It is worth noting that, concerning the number of response options on Likert-type scales, Simms et al. (2019, p. 564) developed a study that compares scales ranging from two to eleven points. Their investigation did not identify improvements in psychometric accuracy on scales above six response options. These authors conclude that "Going beyond six options may confuse participants who perhaps have difficulty perceiving differences between similarly worded response options (e.g. strongly agree vs very strongly agree). Alternatively, more differentiated response scales may pose important challenges to the ability of humans to make fine-grained distinctions regarding responses to relatively coarse psychological test items". Therefore, in the present study, the authors chose to use the same Likert-type scale applied in Souki et al. (2020) study on a la carte restaurant consumers' behaviour, allowing us to compare our results with those obtained in that research.
The platform used for data collection (Google Forms®) has a feature that forces participants to answer all questions indicated as mandatory. As mentioned earlier, all Memorable restaurant experiences affect eWOM questions on the electronic form had predefined answer options, including the "do not know/ not applicable" alternative. Therefore, there are no cases of missing values, typing errors or values outside the limits provided for in the scales used in this survey (Malhotra et al., 2017).
The present research took place in Brazil from 02/03/2020 to 02/28/2020, totalling 25 days that the electronic forms were available on the internet. This survey sample consisted of 475 university students selected using the non-probabilistic snowball technique (Malhotra et al., 2017). These authors point out that non-probabilistic samples can provide reasonable estimates of the characteristics of the population studied, but the results obtained do not allow extrapolation. Thus, the authors of this research asked their students to respond to the survey and disseminate it to other university students in their network. Furthermore, scholars from several Brazilian universities disclosed and requested that their students participate in this study. Therefore, the respondents' participation was voluntary and based on their interest in giving their opinion on the subject addressed in this study.
The researchers chose university students for this study due to the following reasons. The present study used the constructs of perceived quality, positive and negative emotions and propensity to loyalty from the scale proposed by Souki et al. (2020) with university students who are consumers of a la carte restaurants. Thus, the authors sought to adopt a profile of restaurant consumers similar to the study mentioned above. Moreover, casual restaurants in Brazil sell dishes and drinks with prices available separately on the menus so that consumers can choose which combinations best suit their needs and desires. In addition, casual restaurants charge affordable prices to cater to consumers with diverse profiles, such as families, friends and students. Kim et al. (2022) argue that although university students do not represent the totality of consumers of casual dining restaurants, they are one of the customer profiles of this type of establishment. However, this study's results cannot be generalised to all consumer profiles of casual dining restaurants. Finally, Taylor et al. (2012) argue that samples of university students are suitable for this type of research, as they consider one of the profiles included in the population of interest (restaurant consumers, SNS users and that possibly communicate through eWOM). Also, student samples tend to be more homogeneous, favouring theory extraction and reducing the Type II error compared to more heterogeneous samples. Hair et al. (2019a) suggest three criteria to define the sample size. The first refers to the absolute sample size, which should not be smaller than 50 observations and preferably more prominent than 100 observations. This survey had 475 valid cases. The second recommended criterion considers the relationship between the number of observations concerning the number of variables analysed. Generally, five observations are required for each variable TQM included in the model. Considering that the hypothetical model of this research has 79 variables, this study obtained a ratio of 6.01 respondents per variable. This ratio meets the parameter suggested by Hair et al. (2019a). The third criterion is based on the strength of the items' communality that compounds the model's factors. Hair et al. (2019a, p. 133) suggest the "following sample size guidelines: (1) a sample size of 100 is sufficient if all the communalities are 0.70 or above and there are at least three variables with high loadings on each factor; (2) as the communalities fall to the range of 0.40-0.70 then the sample size should be at least 200; and (3) if the communalities are 0.40 or more and there are few high loadings per factor, sample sizes of up to 400 are appropriate". All communalities were more significant than 0.498, the lowest factor loading was 0.714, and the sample had 475 restaurant consumers. Hence, all criteria recommended by Hair et al. (2019a) were met.
The researchers also used the G* Power 3.1.9.4 software (Faul et al., 2009) to assess the sample size and the power of statistical analyses Chin and Newsted, 1999). The construct with the highest number of predictors in the structural model is memorable experiences, which receive an impact (arrows) from three constructs (perceived quality, positive emotions and negative emotions). Considering the memorable experiences predictors, the significance level of 5%, the statistical power of 0.08 and the average effect size (f 2 5 0.15, which is equivalent to a moderate effect of R 2 5 13%), the minimum size of the indicated sample is 107 cases. However, more demanding criteria considering a significance level of 1%, the statistical power of 0.01, and the mean effect size of f 2 5 0.15 indicate that the minimum sample size should be 205 cases. Ringle et al. (2014) suggest doubling or tripling this size to obtain a more consistent model. This research had 475 respondents, which corresponds to 4.44 times more than the least rigorous criterion and 2.32 times more than the most conservative parameter. The post hoc analysis of the G* Power 3 indicated a statistical power of 0.999 for the research model, which suggests that the sample size adopted is adequate.
The hypothetical model of this research was tested using structural equation modelling (SEM). Hair et al. (2019b) and Ali et al. (2018) affirm that the main approaches to SEM are covariance-based SEM (CB-SEM) and partial least squares SEM (PLS-SEM). According to Ali et al. (2018) and Hair et al. (2014), CB-SEM uses a maximum likelihood estimation procedure to reproduce the covariance matrix. On the other hand, PLS-SEM estimates partial least squares based on regression to explain the variance of the unobserved construct, minimising errors and maximising the R 2 values of the endogenous (target) constructs. Ringle et al. (2014) state that the CB-SEM is not indicated when the data do not have a normal distribution or the structural models are complex (many constructs and variables observed). In these cases, the author recommends using PLS-SEM. PLS-SEM involves a twostep evaluation process, one for the measurement model and the other for the structural model . Initially, researchers should assess the reliability and validity of the scale's measurement items (Henseler et al., 2009). Structural model estimates should be made in the second step if no problems are found during the evaluation of measurement items .
SmartPLS analysed this survey's data. Several authors used this software in studies on marketing (Hair et al., 2019b) and restaurant consumer behaviour (Ahmed et al., 2022;Souki et al., 2020). Ringle et al. (2014) recommend using SmartPLS to run complex structural models that include relationships between multiple variables. SmartPLS is ideal when the research data does not have a normal distribution (Hair et al., 2019b). It is worth noting that data that does not show a normal distribution is quite frequent in studies in applied social sciences (Oliveira et al., 2021;Yang et al., 2021). This software can simultaneously estimate the survey's measurement and structural models (Oliveira et al., 2021). SmartPLS also assess a variable's moderation on the effect of an exogenous construct on an endogenous construct (Hair et al., 2014). This research evaluated the moderating effect of CBE-SNS on the relationship between memorable experiences and eWOM.

Exploratory data analysis
During the exploratory analysis of the data, the researchers sought to detect possible problems and solve them to avoid distortions in the results (Tabachnick and Fidel, 2013). The first problem identified was the existence of 38 cases that presented 10% or more responses in the alternative "do not know/not applicable". The researchers considered that such respondents were not adequately qualified to give an opinion on the topics addressed in the present study and opted to discard their answers. Malhotra et al. (2017) recommend eliminating cases where the proportion of unsatisfactory responses is too high to avoid research bias. Thus, after excluding 38 cases (7.41%) from an original sample of 513 elements, 475 cases were kept for further statistical analyses. This number of respondents also meets the parameters suggested by Malhotra et al. (2017) for the elimination of unsatisfactory cases, in which: (1) they can be discarded when they represent less than 10% of the total number of respondents and (2) when the final sample size is large.
After excluding the 38 cases, there were still 796 cells with "do not know/not applicable" responses in a base with 37,525 cells (matrix with 79 columns 3 475 rows), representing 2.12% of the total. There was a high dispersion of "do not know/not applicable" responses among the cases and among the variables of the electronic form, not justifying the exclusion of new cases (Tabachnick and Fidel, 2013). Such responses may come from aspects of restaurant experiences that do not apply to all consumers. The 796 cells with "do not know/ not applicable" responses were then replaced by the arithmetic mean of the responses, as Rubin (2004) recommends.

Description of the sample
The final sample consisted of 475 students from universities in Brazil. Among them, 68.4% are undergraduate students, and 31.6% are postgraduate students. Regarding gender, 60.2% are women and 39.6% are men. Finally, there was a greater concentration of respondents in the 18-30 age group (69.9%).

Method of estimating the measurement model
Confirmatory factor analysis (CFA) tested the measurement model. Initially, the variables that compound each construct were specified. Then, the factor loadings (λ) of the variables of each construct were verified. It is worth mentioning that such loads must be greater than 0.6 (Hair et al., 2019a). The bootstrapping test verified if the factor loadings were significant with a p-value < 0.05. All variables are significantly lower than 0.001.
According to Cronbach's Alpha coefficient (CA), practically all the constructs presented values greater than 0.8, as Hair et al. (2019b) suggested for previously tested scales. Only the construct CBE-SNS had a CA of 0.756, and this indicator can be considered satisfactory for scales under development.
The researchers tested the constructs' convergent validity through the average extracted variance (AVE). This indicator assesses the average percentage of variance shared between the latent constructs (Fornell and Larcker, 1981). The convergent validity is attested when the AVEs of the constructs are more significant than 0.50 . Table 2 reveals that all variables presented indicators above 0.501, except for the global perceived quality (GPQ). It is because such a construct is second order, constituted by the variables of perceived quality of the first-order constructs. TQM DV assesses the degree to which two similar concepts are distinct (Hair et al., 2019a). The first method to test the DV was the criterion of Fornell and Larcker (1981), which checks whether the square roots of the AVEs of each construct are superior to the Pearson correlations between them. Table 3 demonstrates that the square roots of AVE for all constructs were more significant to Pearson's correlations, indicating that the measurement model has DV .
Furthermore, the heterotrait-monotrait (HTMT) ratio of common factor correlations also tested DV, as suggested by Henseler et al. (2015) and Hair et al. (2019a) for PLS-SEM. Ahmed et al. (2022) and Yang et al. (2021) also used HTMT in their studies with structural models using PLS-SEM. The HTMT criterion is the mean value of the indicator correlations across constructs relative to the geometric mean of the average correlations of indicators measuring the same construct (Hair et al., 2019a). Thus, HTMT reliably estimates the true correlation between two constructs. Therefore, high HTMT values indicate problems with DV. Henseler et al. (2015) and Hair et al. (2019b) argue that the HTMT values should be less than 0.90 if the model includes conceptually similar constructs. However, the authors suggest a more conservative value of 0.85 when the constructs are more distinct. The survey's results demonstrate that none of the HTMTs surpasses the recommended standard, and the highest value for the whole model is 0.754 (Table 4). Thus, the DV is adequate for all the model's constructs.
The HTMT2 test was not used in this research because one of the model's constructs (negative emotions) presents negative correlations with other constructs. It is worth noting that Roemer et al. (2021) state that HTMT2 can only be determined if all correlations between the observable variables involved are positive. Hair et al. (2019a) claim that Harman's single-factor test makes it possible to verify the common method bias (CMB) existence. To confirm that the CMB is not present, these authors recommend that the analysis of the main components of the indicators measured in the hypothetical model should represent less than 50% of the variance shared between them in the first eigenvalue obtained. In the present study, Harman's single-factor test included the indicators related to the fifteen constructs that make up the model. The results show that the first factor was responsible for only 28.16% of the explained variance. Therefore, this value    Hair et al. (2019a). Fuller et al. (2016) point out that CMB should be considered a severe problem for research results only when the variance of common methods is effectively high. Finally, Harman's single-factor test is quite robust in identifying common method variance problems if the AVEs and scale reliability meet the reference parameters indicated in the literature (Hair et al., 2019a). Thus, the CMB does not constitute a restriction for interpreting the results of the present study.

Nomological model analysis
Structural models assess whether the measures are valid and reliable and support the hypotheses and relationships between the constructs predicted in theory . Hence, structural models test the causal relationships between constructs through a nomological chain (Hagger et al., 2017), constituting a crucial step in testing the research hypotheses.
The structural model was evaluated by its path coefficients ( §) and significance (α). Path analysis shows the impacts of one construct on the other through arrows that indicate causeand-effect relationships (Hair et al., 2019a). The researchers used the bootstrapping technique, in which several samples are taken from the original data to estimate the model to calculate the § and their significance α (Aguirre-Urreta and R€ onkk€ o, 2018). Figure 2 shows the model paths and their coefficients and significance.
The model's Pearson coefficient of determination (R 2 ) is another indicator to be observed. It allows evaluation of the portion of the variance of the endogenous variables explained by  Ringle et al. (2014) state that the value of R 2 has a negligible effect when it is equal to or less than 2%, has a medium effect when it is equal to 13% and has a strong impact when it reaches a percentage greater than or equal to 26%. Figure 2 presents the R 2 values of the constructs evaluated in this research's structural model. This research's hypotheses refer to the impacts of the GPQ on memorable experiences (H1) and positive emotions (H2), and negative emotions (H3). The GPQ impacted memorable experiences ( § 5 0.242), positive emotions ( § 5 0.599 and R 2 5 35.9%) and negative emotions ( § 5 À0.229 and R 2 5 5.3%). Shahzadi et al. (2018) and Marinkovic et al. (2014) also found that GPQ impacted consumers' positive emotions. As in the present study, Jung et al. (2021) also found that increasing GPQ reduces consumers' negative emotions. The GPQ, associated with positive emotions ( § 5 0.431) and negative emotions ( § 5 0.07), helped explain consumer memorable experiences (R 2 5 35.6%). These results support the five hypotheses regarding memorable experiences (H1-H5).
This study's results demonstrated that memorable experience significantly impacts the propensity to loyalty ( § 5 0.535 and R 2 5 28.6%), confirming H6. This result is consistent with previous studies that demonstrate that memorable experience is an antecedent of propensity to loyalty (Rasoolimanesh et al., 2021;Ali et al., 2014). Finally, this research's structural model confirmed that memorable experiences impacted eWOM ( § 5 0.365), with a significance of 0.001, supporting H7.

Moderating effect of consumers' behavioural engagement in the SNS on the relationship between memorable experience and eWOM
The researchers tested two models to assess the moderating effect of CBE-SNS on the relationship between their memorable experiences with casual dining restaurant experiences and eWOM (H8), including or excluding the moderating construct. Moderation describes a situation in which the relationship between two constructs is not constant but is dependent on the values of a moderating variable. Such a variable modifies the strength or direction of the relationship between two constructs in a structural model .
This survey's results reveal that the R 2 of eWOM without the construct CBE-SNS was 17.8%. On the other hand, when the hypothetical model includes the construct CBE-SNS, the R 2 increases to 31.1%. Figure 2 shows that the moderator construct has an effect (f 2 ) of 0.162 with a significance of 0.01 (two-tailed) in the relationship between memorable experiences and eWOM. Hair et al. (2017) argue that this effect is significant for moderating relationships: (1) f 2 5 0.005 -small; (2) f 2 5 0.010 -average; (3) f 2 5 0.025 -large. This result confirms H8 by demonstrating that the CBE-SNS strongly moderates the relationship between memorable experiences in casual restaurants and eWOM. Figure 3 represents the bidirectional interaction effects for standardised variables. According to Dawson (2014), bidirectional interactions demonstrate how a relationship occurs between an independent variable (X) and a dependent variable (Y), moderated by a third variable (M). In the present study, the independent variable is satisfaction. The dependent variable is eWOM. The bidirectional interactions graphically demonstrate that CBE-SNS moderates the relationship between memorable experiences and eWOM. Gardner et al. (2017) state that a variable M can moderate the relationship between X and Y, making it stronger or weaker, depending on its variation. The blue line (Figure 3) represents the relationship between memorable experiences and eWOM when CBE-SNS is high (one standard deviation above the mean). The impact of memorable experiences on eWOM is more pronounced and positive for consumers who are highly behaviourally engaged in SNS. On the other hand, the red dotted line represents the relationship between memorable experiences and eWOM when CBE-SNS is low (one standard deviation below the mean). The low CBE-SNS reduces the strength of the relationship between memorable experiences and eWOM. It is worth mentioning that consumers with a low level of Memorable restaurant experiences affect eWOM behavioural engagement in the SNS present a more passive posture (Bailey et al., 2021), not seeking information or sharing their experiences but only observing content shared by their contacts on social networks (Kanje et al., 2020;Correia et al., 2018). Therefore, consumers with this profile tend to communicate little through eWOM, even when they have memorable experiences in casual dining restaurants.
On the other hand, customers with high CBE-SNS tend to participate more actively in online communications to influence their peers, share their experiences and seek information (Bailey et al., 2021;Gvili and Levy, 2018;Correia et al., 2018). Thus, consumers who have had memorable experiences and high CBE-SNS tend to generate and post content on their lived moments in casual restaurants through eWOM.

Conclusions and academic contributions
The literature has several studies that used the S-O-R theory to understand the behaviour of food service consumers (Lim et al., 2022;Oh and Kim, 2021;Brewer and Sebby, 2021;. However, no previous research has demonstrated the direct impacts of perceived quality by casual restaurant consumers (stimulus) on positive emotions, negative emotions and memorable experiences (organism) and the implications of memorable experiences on the propensity to loyalty and eWOM (response) using such theory. In addition, no previous research has shown a moderating effect of CBE-SNS on the relationship between memorable restaurant consumer experiences and eWOM. Such moderation occurs because people with low CBE-SNS communicate poorly through eWOM, regardless of whether they have had memorable experiences in casual restaurants. On the other hand, high CBE-SNS intensify eWOM only when they retain experiences in memory (Figure 3). Therefore, this study is unprecedented and contributes to the theory by concomitantly proving all the hypothetical relationships between the constructs mentioned above. This research reveals that when consumers positively evaluate the quality of their restaurant experiences, they tend to agree that they felt positive emotions and disagree that they felt negative emotions. Furthermore, it confirms previous studies demonstrating the need to include separate items to measure consumers' positive and negative emotions (Oliveira et al., 2022;Jung et al., 2021;Leung and Wen, 2021;Souki et al., 2020). Positive and negative emotions contribute to the explanatory power of consumers' memorable experiences and, consequently, to their future repercussions, such as the propensity to loyalty and eWOM. These results provide answers about the attitudinal and behavioural consequences of negative emotions, contributing to filling the gap identified by Song and Kim (2021) and Song and Qu (2017). These authors argue that research on restaurant consumer behaviour has focused mainly on positive emotions. Therefore, the present study contributes academically by unravelling some attitudinal and behavioural unfoldings of negative emotions experienced by consumers of casual restaurants.
Another academic contribution of this study is to demonstrate that the mere transposition of research results on WOM to eWOM is not always valid because consumers must first strongly engage with SNS to communicate through eWOM. Therefore, future studies on the direct and positive impacts of memorable experiences on eWOM should consider the influence of CBE-SNS. This research finding suggests that future investigations should include variables that may moderate the relationship between stimulus and organism and between organism and response, enhancing the applicability of the S-O-R theory.

Conclusions and managerial contributions
This study contributes managerially by revealing intangible and tangible dimensions of perceived quality that contribute to the experience of consumers in casual restaurants. Among the intangible dimensions, the service quality, the atmosphere and the customer orientation stood out. The most relevant tangible dimensions for the quality perceived by consumers about their experiences in casual restaurants were infrastructure and food quality. The tangible and intangible dimensions mentioned above constitute crucial elements for assessing the quality perceived by consumers. However, other intangible (reputation, status and social endorsement) and tangible (accessibility and convenience) dimensions also positively impacted consumers' perception of quality, which may represent significant competitive differentials for the value proposition of casual restaurants. Finally, it is worth noting that the results regarding the perceived quality dimensions align with the findings by Souki et al. (2020).
Another contribution of this study is to demonstrate that casual restaurants should offer their customers experiences that generate a high perception of quality, positively impacting their positive emotions, memorable experiences and propensity to loyalty. However, memorable consumer experiences at casual restaurants do not necessarily imply positive eWOM. Thus, restaurants must assess whether the CBE-SNS is low or high and use different strategies for each public behavioural profile. If CBE-SNS is low, the casual dining restaurant should initially encourage consumers to interact more in the SNS to report their memorable experiences through eWOM later. On the other hand, if the CBE-SNS is high, casual dining restaurants should spur consumers to share their memorable experiences through eWOM and monitor whether the reviews are favourable or unfavourable. It is worth mentioning that negative emotions are also retained in consumers' memory, triggering unfavourable eWOM. In this case, casual restaurants should use service recovery strategies to improve the quality of consumer experiences, generating favourable affective, cognitive and behavioural repercussions. Finally, restaurant managers must maintain interactivity, transparency and assertive communication with consumers in all cases.
Another contribution of the present study is that casual restaurant managers can use this survey's questionnaire to monitor perceived quality, positive and negative emotions, Memorable restaurant experiences affect eWOM memorable experiences, the propensity to loyalty, CBE-SNS and eWOM. This monitoring provides subsidies to develop strategies and actions to continuously improve the quality of customers' experience and verify the evolution of its impacts in the affective, cognitive and behavioural dimensions. Managers of different types of restaurants can adapt this survey questionnaire to supply the peculiarities of their business.

Research limitations and suggestions for future research
This study has some limitations. This research included university students from Brazil, one of the casual restaurants' consumer profiles. Hence, this study's results cannot be generalised to the restaurant consumer population, who may have different profiles. Thus, future studies may find different results if they consider other consumer profiles of casual restaurants. In addition, further research may discover different consumer profiles elsewhere. Thus, testing the proposed model in other regions and countries is recommended.
Another limitation of this study is that the sampling system was non-probabilistic by snowball (Malhotra et al., 2017). Thus, the total number of university students invited to participate in the survey is unknown, making it impossible to calculate the response rate and the non-response error and extrapolate its results to the population. Future studies may use probabilistic samples.
As mentioned in this study's methodology, the students answered the electronic forms just before the beginning of the COVID-19 pandemic in Brazil. Thus, another limitation of the study is that the participants responded only about one item related to hygiene in the infrastructure construct: "It has a clean and hygienic environment (bathroom/lounge/tables/ outdoor area/kitchen)". In this context, future studies may include other indicators to assess consumers' perceptions of cleanliness, hygiene and food security in restaurants deeply.
This study contemplated only casual restaurants. Thus, future research may adjust this study's model to other types of restaurants. Among them, Souki et al. (2020) highlight (1) a la carte restaurants (food and drinks available on a menu with individual prices for each item); (2) buffet or self-service restaurants (where consumers pay a fixed price and can have their meal in an all-you-can-eat system); (3) dish-made (restaurants that offer a standardised and single type of dish, ready-to-eat, usually at reduced prices and without additional services); (4) rodizio (Brazilian or Portuguese restaurant that serves a wide variety of grilled and roasted meats served by waiters at tables, with a fixed price charged for an unlimited amount of food); (5) food-per-kilo (self-service system in which consumers can serve their food according to their wishes, but payment is made in proportion to the weight served, considering a fixed price per kilo); (6) fast food chains (restaurants that serve food quickly after ordering and without additional services), among others (Lim et al., 2022;. Finally, the respondents filed the questionnaires at a specific time (single cross-sectional). Future studies may use longitudinal or multiple cross-sectional to monitor the evolution of consumer behaviour over time.