I
#wheretoeat
The impact of electronic word-of-mouth in social
network media on millennials’ purchase intention in
the restaurant setting
Master’s Thesis 30 credits
Department of Business Studies
Uppsala University
Spring Semester of 2017
Date of Submission: 2017-05-30
I
Abstract
Social media has been growing constantly during the past years and social networking is one of the most popular online activities. This development has changed how individuals communicate with each other as it enabled valuable improvements for electronic word-of-mouth. Individuals share experiences, evaluations or opinions regarding products or services within their network, which has driven marketers’ attention towards this tool. Here, millennials are active users of social media, they have a high purchasing power and enjoy spending money on eating out. This study investigated how different determinants of electronic word-of-mouth in social network media impact millennials’ purchase intentions within the restaurant setting. A snowball sampling with 162 respondents was used in order to test seven hypotheses. Results indicated that attitude towards information usefulness and subjective norm are influencing millennials purchase intention. However, this was not the case for perceived risk. Managers within the restaurant setting are advised to consider that the quality and the source of a post generate a positive attitude. Posts with a high awareness increase social pressure to visit a certain restaurant. This will in turn lead towards a positive impact on millennials’ final purchase intention and managers need to focus on these strategies when targeting this generation. Keywords
II
Acknowledgements
First of all, we would like to express our gratitude to our thesis supervisor, Anna Bengtson, for her valuable insights. We would also want to thank James Sallis for his constructive comments. Without the comments, insights and guidance throughout this process by Anna, James and our colleagues, this paper would not have been the same. A special thanks goes to all respondents for choosing to be a part of this study. Last but not least, we want to thank people that are close to us, for their support and encouragement.
Vendela Tersén and Katharina Wecken, Uppsala, 2017-05-30
III
Table of Content
List of Tables ... IV List of Figures ... IV List of Abbreviations ... V 1 Introduction ... 1 1.1 Problem Formulation... 3 1.2 Research Purpose ... 4 1.3 Outline ... 5 2 Theory ... 6 2.1 Theoretical Setting ... 6 2.1.1 Millennials ... 62.1.2 Restaurants as a Service Setting ... 7
2.1.3 Online Customer-to-Customer Interaction ... 7
2.1.4 Electronic Word-of-Mouth ... 8
2.1.5 Social Network Media ... 9
2.1.6 Electronic Word-of-Mouth in Social Network Media ... 10
2.2 Theoretical Framework ... 11
2.2.1 Purchase Intention ... 11
2.2.2 Theory of Reasoned Action ... 12
2.2.3 Information Adoption Model ... 14
2.2.4 Social Interaction ... 15 2.2.5 Perceived Risk ... 16 2.2.6 Summary ... 17 3 Method ... 19 3.1 Research Design ... 19 3.2 Research Strategy ... 19 3.3 Sampling Frame ... 20
3.4 Measurements and Scales... 20
IV
4.1 Sample Characteristics and Data Cleaning ... 26
4.1.1 Sample Characteristics ... 26
4.1.2 Cleaning Data ... 26
4.1.3 Outliers ... 26
4.2 Construct Validity and Reliability... 27
4.3 Normality ... 30 4.4 Hypotheses Testing ... 31 5 Analysis ... 35 6 Discussion... 38 6.1 Summary ... 41 6.2 Managerial Implications ... 41
6.3 Limitations and Further Research ... 43
Reference List ... 45
Appendix 1 ... 54
List of Tables
Table 1: Hypotheses ... 18Table 2: Constructs and Items... 23
Table 3: Rotated Component Matrix - Factor Analysis 1 ... 28
Table 4: Rotated Component Matrix - Factor Analysis 2 ... 28
Table 5: Rotated Component Matrix - Factor Analysis 3 ... 29
Table 6: Cronbach's Alpha ... 30
Table 7: Test of Normality - Skewness and Kurtosis ... 31
Table 8: Multiple Regression IAM ... 33
Table 9: Multiple Regression Social Interaction ... 33
Table 10: Multiple Regression TRA and Perceived Risk ... 34
List of Figures
Figure 1: Theory of Reasoned Action ... 13Figure 2: Information Adoption Model ... 15
Figure 3: Social Interaction ... 16
Figure 4: Conceptual Model (Gunawan and Huarng, 2015) ... 18
V
List of Abbreviations
AT Attitudes towards information usefulness
AQ Argument quality
CCI Customer-to-customer interaction
eWOM Electronic word-of-mouth
IAM Information Acceptance Model
KMO Kaiser-Meyer-Olkin
PI Purchase intention
PR Perceived risk
SC Source credibility
SN Subjective norm
SNM Social network media
SINF Social influence
SINT Social integration
TRA Theory of Reasoned Action
VIF Variance inflation factor
1 Introduction
Social media has been growing constantly over the last years (Statista, 2016). During the last six years, social media has gained around 1,3 billion new members, reaching a number of 2,34 billion users worldwide in 2016. A further growth of 30% is expected during the upcoming years, reaching a number of 3 billion users by 2020 (ibid). As social networking is one of the most popular online activities, it is nowadays an established part of peoples’ everyday life, especially among the younger generations. Here, individuals grew up with this technology and can not imagine life without social network media. Facebook, Twitter and Instagram changed the way individuals communicate with each other. Yet, they do not only use social media to stay in touch with friends but also as a search tool for product and service information, to express opinions or simply for entertainment. This all represents the high user engagement of social network media (Erkan, 2016; Statista, 2016).
Social network media (SNM) is defined as Internet-based applications that allow individuals to create and exchange user-generated content (Boyd and Ellison, 2007). This content is created by different facilities that are provided by the social media website, like pictures or videos and can contain personal information as well as product related content. In fact, SNM recently started to be considered as a great opportunity to express and share product- or service-related experiences and opinions (Chu and Kim, 2011; Statista, 2016). This, in turn, can create attitude changes about the discussed product or service in peoples’ minds, leading towards an impact in their potential purchase behavior. Therefore, marketers pay increasing attention to SNM as a tool to address customers and strive to establish an own successful SNM-presence as well as to engage customers sharing their (positive) experiences with a product or service, to potentially inspire others towards a purchase.
However, the characteristic of receiving recommendations from people you know and trust got replaced by an anonymous environment, where readers of electronic word-of-mouth (eWOM) do not know how much they can trust the source with the given information. Here, SNM returned the original characteristics of WOM, where people are connected to the sender of the message since they are “friends” or follow people on their SNM account (ibid). This reduced anonymity offers the potential to make eWOM more trustworthy and reliable than traditional WOM and by this the potential to have an even greater impact on the reader and its purchase intention. Indeed do conversations in SNM frequently refer to products or services and are therefore naturally influential on customers purchase intention. In consequence, it offers marketers a huge potential to reach customers within their personal environment and spread information with a higher power and conviction that drives them towards purchasing that product or service (Chu and Kim, 2011; Soares, Pinho and Nobre, 2012). Since customers can not try and return a service to the same extent as it is possible with most products, receiving sufficient information based on other’s experiences is a good foundation to get an impression about the service quality and to feel confident with making future purchases (Grönroos, 1984). Hence, knowledge about how eWOM in SNM works should be particularly of importance for marketers in the service setting.
Recent studies stated that users increasingly apply SNM for the purpose of getting information about unfamiliar products and services, emphasizing that posts or comments by others are taken into account to create an attitude about a product and service, which in turn impacts the purchase intention (Mortazavi, Rahim Esfidani and Shaemi Barzoki, 2014; Soares et al., 2012). Based on the reduced anonymity in combination with the still existing speed of information diffusion, eWOM on SNM is seen as a powerful marketing tool, since users let the posted information influence their behavior (Chu and Kim, 2011). However, similar to the concepts of WOM or eWOM, not all information have a strong impact on users’ purchase intentions (Erkan and Evans, 2016; Matute et al., 2016). It is therefore of importance which determinants are powerful within the SNM setting when it comes to persuading its users towards a purchase and how strong their relationship is.
within the technological environment. Different to the generations before, they embrace new interactive media as SNM and are not only active users (Moore, 2012), they also have a high purchasing power (Schewe et al., 2013), which makes them an interesting market segment for companies who are active in SNM. One activity millennials value a lot is their lifestyle, especially concerning their nutrition and eating out. They eat out more often than other generations (3,4 times vs 2,8 times per week) and not only visit restaurants or cafés to simply eat something but rather see it as a group activity and a way to express their self (Barton, Koslow, Fromm and Egan, 2012). In addition, they enjoy exploring new restaurants and cafés; 40% order a different meal each time they eat out (Millennial Marketing, 2017).
Hence, millennials become an interesting target group for marketers, as they are open for new advice, especially by their friends. Millennials Marketing (2017) found out, that 68% of millennials seek out friend’s opinions before they decide for a restaurant. Looking at SNM, sharing the experience of a restaurant visit or food consumption is an ongoing trend. On Instagram, #food is one of the most popular hashtags. Here, 90 new photos hashtagged with #foodporn are uploaded every minute (Menulog, 2014). Especially millennials are prone to share their pictures on SNM, letting their network know where and what they eat (Marchak, 2016). Therefore, it is a logical consequence, that SNM can influence millennials’ decision of
which restaurant1 they want to visit. Understanding the procedures of how different
determinants of eWOM in SNM are impacting their purchase intentions can be of high importance for marketers. A profound knowledge could help to actively stimulate SNM postings with beneficial content and impact customers’ decision-making process. This knowledge can not only help marketers to target millennials, but also to establish a good market position within an environment that faces increasing competition where a vast range of organizations vie for the customers’ awareness and loyalty.
1.1 Problem Formulation
SNM and eWOM are powerful in influencing individual’s decision making as conversations among SNM users often refer to products or services and create certain attitudes about a specific product or service (Chu and Kim, 2011). However, not all information need to be influential. Users face a huge amount of information on SNM and therefore critically screen
the information before using and internalizing it (Erkan and Evans, 2016). In other words, simply reading about restaurants on SNM does not automatically influence the user’s attitude or their purchase intention. Here, the determinants for the mechanism between eWOM on SNM and the customer’s purchase intention have not yet been fully discovered and researchers as well as marketers do not know how factors turn SNM users towards a potential purchase. Previous literature in this field has focused on either the characteristics of eWOM information (Cheung, Lee and Rabjohn, 2008) or the customer’s behavior towards this information (Reichelt, Sievert and Jacob, 2014; Wang et al., 2012). Nevertheless, according to Knoll’s (2016) review of recent eWOM studies, the attention should be focused on both, message and customer. This is in line with the research Erkan and Evans (2016) recently conducted, combining both elements.
Therefore, a consideration of both aspects is necessary when investigating the impact of eWOM in SNM. Furthermore, a specification concerning the target group and the industry would enable a more specific understanding of the phenomenon. Even though Erkan and Evans (2016) did consider the characteristics of eWOM information as well as the customer behavior, they did not have further restrictions concerning the setting and therefore kept their findings rather broad. Limiting research towards a certain setting offers more valid understanding as different industries have different mechanisms. As millennials not only are familiar with SNM but also seen as the generation with the highest purchasing power, it seems apparent that they are the generation of interest. In addition, restaurants are a fast growing industry where competition is increasing (Han, Nguyen and Simkin, 2016). Managers within the restaurant setting need to overcome this hostile and competitive environment by understanding what influences customers purchase intention. By knowing how different determinants impact millennials, managers could more precisely drive their SNM marketing activities and reach a promising target group more efficiently. Hence, a combination of a powerful generation that spends a lot of money with an industry they perceive as highly relevant would offer a high potential to increase revenues for managers. Therefore, the industry of restaurants is worth investigating.
1.2 Research Purpose
not only of interest which determinants impact purchase intention but also the intensity of the relationship, in order to investigate how the determinants need to be considered within organizations’ social media marketing.
This leads towards the following research question:
How do different determinants of electronic word-of-mouth in social network media impact millennials’ purchase intention within the restaurant setting?
To investigate this question, a combination of Theory of Reasoned Action (TRA), Information Acceptance Model (IAM) and perceived risk is used in order to cover traditional and well-used models to predict purchase intention but further complemented with modern elements of the SNM. This combination of models has been used before by Gunawan and Huarng (2015) within the context of SNM sites’ viral effect on customer’s purchase intention. For this study, the setting is changed towards eWOM in SNM, since this is, as already explained before considered a powerful marketing tool (Chu and Kim, 2011). Furthermore, adaptions have been made for the target group of millennials and the restaurant setting.
By answering this research question, the study results in both practical and theoretical understanding regarding the impact of eWOM in SNM towards individual’s purchase intention. On the theoretical side, the combination of the used models can be strengthened or further developed. On the managerial side, a better understanding can help marketers to improve their SNM marketing activities within the restaurant setting to target millennials.
1.3 Outline
2 Theory
2.1 Theoretical Setting
2.1.1 Millennials
Generational cohort theory proposes that individual’s distinct behavioral and psychological profiles are created from major historical and societal changes in the individuals’ formative years (Strauss and Howe, 1991). A generational cohort is not defined by the year of birth but
rather of external events that individuals experienced and could, therefore, be more effective
when segmenting a market (Schewe, Meredith and Noble, 2000). Millennials are a generational cohort, as they are a group of individuals that share similar experiences and unique characteristics (Beldona, Nusair and Demicco, 2009). There is not an exact age-range that defines the millennial generation, but individuals are born around 1980-2000 (Gurău, 2012; Tyler, 2007).
Millennials are a large generation of today (Nowak et al., 2006; Schewe, Debevec, Madden, Diamond, Parment and Murphy, 2013). They expect to be heard and are often overconfident (Allsop, Bassett and Hoskins, 2007). However, if given criticism they feel uncomfortable (Tyler, 2008). Millennials struggle with independent decision making (Howe and Strauss, 2007; Tyler, 2007) and the fear of making mistakes (Howe and Strauss, 2007). Therefore, they often ask parents for advice (Tyler, 2007). Further characteristics of this generation are the urge for instant gratification, receiving information immediately and craving for feedback. This is also a reason for why millennials can be perceived as high maintenance according to Tyler (2007).
2013), it is critical to understand these consumption patterns (Gurău, 2012).
Millennials are food fanatics (Turow, 2015). Restaurant visits are perceived as a group activity, and this generation experiences a social value of eating out (Barton et al., 2012). When comparing with other generations who eat out 2.8 times per week, millennials eat out 3.4 times a week (ibid). Millennials desire to try new food, both local and international cuisine (Turow, 2015) as they feel the need of exploring something new in their choice of restaurants (Barton et al., 2012). Considering millennials digital presence, purchasing power and restaurants visits, it is from a managerial perspective important to understand how eWOM in SNM influences this generation’s purchase intention.
2.1.2 Restaurants as a Service Setting
The restaurant industry is growing (Binkley, 2006). Changes in customers’ lifestyles and family structures (Warde, Cheng, Olsen and Southerton, 2007) have led to a growing habit of eating out (de Rezende and de Avelar, 2012). However, restaurants do no longer only supply food (Chan, Berger and Van Boven, 2012) and functional benefits such as food quality (Han et al., 2016). Meals at restaurants are normally consumed with other individuals to maintain relationships and fulfill social needs (Rozin, 1996). Restaurants similarly provide symbolic benefits as social status and group membership can be demonstrated by visiting certain restaurants (Han et al., 2016; Warde et al., 2007; Witt, 2010). Specific restaurant visits can through identification express social identity and belonging to a social group (Wattanasuwan, 2005). If a restaurant is socially accepted by others, customers who consume this service will form their identity (Han et al., 2016).
Reasons for visiting restaurants can be many, nonetheless, customers need to search information to have the possibility to compare existing alternative before their purchase decision (Pedraja and Yagüe, 2001). As restaurants are the fastest growing service industry and competition is increasing (Han et al., 2016) managers need to overcome this hostile environment by understanding what influences customers purchase intention.
2.1.3 Online Customer-to-Customer Interaction
The term CCI can be described as “the transfer of information from one customer (or a group of customers) to another customer (or group of customers) in a way that has the potential to change their preferences, actual purchase behavior, or the way they further interact with others” (Libai, Bolton, Bügel, de Ruyter, Götz and Risselada, 2010, p. 269). This definition highlights that the information is created by, and presented from the perspective of the customer and can include experiences, evaluations or opinions about a product or service (Park, Lee and Han, 2007). Content created by the customer is perceived to be more trustful and credible than company generated content with a greater persuasiveness of the given information (Bickart and Schindler, 2001). This is due to the assumption that customers do not have the interest to manipulate the reader in order to make them purchase. It is furthermore expected to reflect the typical product performance. Different research has proven, that CCI influences customers’ purchase intention since individuals learn from and are affected by other opinions or behaviors (Chen et al., 2010; Godes and Mayzlin, 2004; Park et al., 2007). Here, it can be differentiated between learning by observing others behavior and imitate it, as well as learning by creating an own opinion based on reading or listening to other customers’ opinion and experiences (Chen et al., 2010; Libai et al., 2010). Learning and observing can take place based on text, pictures or videos (Chen et al., 2010). In other words, people can watch others visiting a restaurant and post about it online, where the desire arises to visit the same place or they can read peoples’ posts, directly recommending a restaurant, that turns the reader towards doing as recommended.
Originally, the term WOM was used to describe CCI. However, with the increasing diversity of CCI in the online environment, a broader view of this concept is needed (Libai et al., 2010). Nevertheless, WOM or eWOM is still one important aspect within CCI and will be described in the next section.
2.1.4 Electronic Word-of-Mouth
websites, newsgroups, social media, blogs or shopping websites (Cheung and Thadani, 2012; Hennig-Thurau et al., 2004). As eWOM is one element of CCI, it also influences customers’ purchase intention which is argued by various articles (Park et al., 2007; See-To and Ho, 2014). The fundamental idea of eWOM is similar to traditional WOM. However, the development from traditional WOM towards eWOM allowed customers to not only perceive information from people they know, but rather from a vast, geographically spread group of people, who experienced the product or service of interest (Cheung et al., 2008; Ratchford, Talukdar and Lee, 2001). Whereas WOM used to be face-to-face within private conversations, eWOM takes place within a computer-mediated context, where the environment is anonymous (Zhang, Craciun and Shin, 2010). Consequently, social cues are lacking and the credibility of the sender is unknown, as the reader has to rely solely on the content of the message. Here, similarity, quality and quantity of reviews are perceived as determinants for its trustworthiness and for the customer’s responses (Matute et al., 2016; Reichelt et al., 2014).
For the customer, the increased volume of eWOM enables to reduce risk and uncertainty before making a purchase by collecting more information about the product of interest (King et al., 2014). In addition, prior research has shown, that customers see eWOM as more persuasive than traditional marketing media, like TV or radio advertising (Cheung and Thadani, 2012). Since the online environment offers the chance to spread a message quickly and accordingly reach a vast number of people, it enables the opportunity to create bigger awareness of a product or service across communities or media (King et al., 2014). In consequence, eWOM offers the opportunity to affect firm-level outcomes, such as product sales, revenues or stock prices (Chevalier and Mayzlin, 2006).
2.1.5 Social Network Media
connections with strangers based on shared interests, views or activities (ibid). SNM can be used for both private and professional interaction (Trusov, Bucklin and Pauwels, 2009). Moreover, the intention to use SNM can vary and include information search, communication, social interaction or entertainment (Mortazavi et al., 2014).
To enable people to interact, users create a personal profile, where they present information about their age, location or interests (Boyd and Ellison, 2007; Erkan, 2016). In addition, users can upload a profile picture and further personal pictures, videos or other content. The visibility of these profiles varies amongst different platforms (Boyd and Ellison, 2007). Furthermore, on SNM people can identify and connect with other users. Most of the relationships are bi-directional, where the connection must be confirmed by both sides (often claimed as “friends” or “contacts”), whereas some others have a one-directional relationship, where terms like “fans” or “followers” are applied (ibid). Each connection creates content, which is likely to be valuable for the user. For many people are SNM an established part of their everyday life (Trusov et al., 2009).
SNM offers the possibility to share product information, opinions and experiences, for example by writing own reviews, sharing articles or comment on posts. It is even possible to reach a global audience who share the same interests in a product or service (Chu, 2009). This, in consequence, has lead towards changes in the search for product information and making purchase decisions, where people often actively use SNM to gather information presented by other customers. In addition, it has driven the attention of marketers towards SNM as they try to use them to their advantage. SNM are not only interesting for marketers because they allow reaching a large number of people, they also provide a very specific targeting of the customer (Gironda and Korgaonkar, 2014). Companies can have their own SNM-profile and better communicate with their customers as well as engage them to interact with current and potential
customers. Hence, SNM provide a novel way to build and maintain customer relationships.In
addition, users can pass along marketer generated content and share their own thoughts (Chu, 2009).
2.1.6 Electronic Word-of-Mouth in Social Network Media
they have got the possibility to debate issues within their existing network (Erkan and Evans, 2016), meaning they have some kind of relationship towards the person sharing, writing or commenting on information. The reduced anonymity makes eWOM more trustworthy and reliable, as it includes the source and not only the content (ibid). Based on the customer generated content and the still existing ease and speed of information diffusion, Soares et al. (2012) describe SNM as a powerful tool for eWOM, where users can spread their product or service-related experiences and opinions freely among other users. Chu and Kim (2011) even state, that eWOM may become the most powerful source of information with an interpersonal influence.
EWOM in SNM takes place as people can share written text, pictures, videos or applications and by that facilitate the dissemination of information among a huge amount of people (Erkan and Evans, 2016). Just forwarding a former post, created by another individual or a marketer, leads towards sharing thoughts with others and towards a potential impact of the reader’s attitude and purchase intention (Chu and Kim, 2011). Sometimes, the intention of posting does not need to have a product related or advertising purpose – posting product included content, for example, a picture where the user simply consumes a product or service can lead towards indirect eWOM (Erkan and Evans, 2016). From the reader’s point of view, this highlights that eWOM in SNM can be actively searched for as a source of information as well as passively received without being aware of (Chu and Kim, 2011).
By spreading information on SNM, an advantage is taken by the multiplication of messages to transmit it to thousands of people which could at best lead towards a digital hype – an outcome marketers try to achieve with their marketing activities (Gunawan and Huarng, 2015).
For this study, eWOM in SNM will cover all types of occurrence, from indirect postings with product included content to customer-generated information about a product or service with a promotional intention. In addition, the study will not focus on a specific SNM but rather address all different types and platforms of SNM.
2.2 Theoretical Framework
2.2.1 Purchase Intention
2009; Foxall, 2005; Infosino, 1986; Morrison, 1979; Sun and Morwitz, 2010). External factors, as social-economic circumstances, demographics and in-store stimuli, might influence the final purchase decision (Ajzen and Fishbein, 1980). In addition to previous experiences, beliefs and attitudes (Simonson and Rosen, 2014), eWOM have become a significant information source
and influence customers purchase intention (Chevalier and Mayzlin, 2006; Schindler and
Bickart, 2012; Senecal and Nantel, 2004). When customers purchase they take into consideration the nature of the product (Holton, 1958), the level of knowledge about the alternatives (Sussman and Siegal, 2003) and the perceived risk of the product (Bauer, 1960). Theory of Reasoned Action (TRA) is a well-used model to analyze purchase intention and this model is therefore appropriate to use for the purpose of this study. The TRA will be complemented by the Information Adoption Model (IAM) to fully analyze purchase intention in the context of eWOM. IAM has been used in previous research in the context of eWOM and purchase intention (Cheung et al., 2008; Erkan and Evans, 2016). Perceived risk will also be taken into consideration as another influencing factor for purchase intention. In this study, a model combining TRA, IAM and perceived risk will be used. This combination has earlier been tested in Gunawan and Huarngs (2015) study where the focus was on viral effects in social media on consumers’ purchase intention. Based on the similarities of their setting and this study’s environment, the usage of this combination is perceived to be appropriate. The next section introduces the theoretical model with its modifications.
2.2.2 Theory of Reasoned Action
Theory of Reasoned Action (TRA) investigates behavior and it argues that individual’s attitudes do not always need to lead to a specific behavior, it is rather a determinant of individuals’ behavioral intention (Ajzen and Fishbein, 1980; Fishbein and Ajzen, 1975). In this study, behavioral intention is addressed as purchase intention to avoid vocabulary misunderstanding. In previous research, the TRA model has been frequently used to measure the relationship between eWOM and purchase intention (Cheung and Thadani, 2012;
Prendergast, Ko and Yuen, 2010; Reichelt et al., 2014). Two components of the TRA model
that are influencing purchasing intention are attitude and subjective norm (Ajzen and Fishbein, 1980) and these will be used in this study (see figure 1).
consciously consider the behavior with the most desirable outcome (Ajzen and Fishbein, 1980). In SNM other customers write messages and from this, individuals learn attitudes and form purchase intentions (Wang et al., 2012). Hence, the interaction that takes place, strongly
influence individual’s attitude towards products and services (Moschis and Moore, 1979).
Subjective norm is how individuals perceive a social pressure of what other people want them to do, whether to perform or not perform a certain behavior (Ajzen and Fishbein, 1980). As an example, technological acceptance is a subjective norm because other people expect technology to be used, which will influence the behavioral intention (Bhattacherjee and Sanford, 2006). It has been argued that subjective norm does not capture intentions of behavior if customers do not have an impact on other customers’ thoughts (Miller, 2002). However, as SNM are social in their nature and individuals interact and influence each other, subjective norm is useful when predicting purchase intentions (Gironda and Korgaonkar, 2014).
There has been criticism regarding purchasing behavior and TRA, as individuals’ incomes can change and unexpected promotions of products might impact customers to not purchase products even if having purchasing intentions (De Cannière et al., 2009; Foxall, 2005; Infosino, 1986; Morrison, 1979; Sun and Morwitz, 2010). External factors as demographics, traditional attitudes and personality traits might influence purchase intention (Ajzen and Fishbein, 1980). In this study, the aim is to investigate purchase intention and not the final purchase decision.
Figure 1: Theory of Reasoned Action This leads to the following hypotheses:
H1: Millennials’ attitude towards information usefulness positively affects the purchase intention of SNM marketed restaurants.
2.2.3 Information Adoption Model
Individuals use eWOM to transfer and receive information (Bansal and Voyer, 2000), yet can the same content impact individuals differently (Cheung et al., 2008). Information adoption process is used in order to understand how people internalize the information received in the knowledge transfer (Nonaka, 1994). According to Sussman and Siegal (2003), TRA does not fit for information adoption since it does not answer the type of questions relevant for the argument quality and source credibility.
Information Adoption Model (IAM) is an integration of the Technology Acceptance Model (Davis, Bagozzi and Warshaw, 1989) and the Elaborated Likelihood Model (Petty, Cacioppo
and Goldman, 1981; Petty and Cacioppo, 1986) with dual-process models of information
influence, to explain how people are influenced in adopting ideas, knowledge or information
(Sussman and Siegal, 2003). In computer-mediated communication contexts, IAM explains
how individuals are influenced to adapt to the posted information (ibid) and this is suitable for studies with an interest of eWOM (Cheung et al., 2008; Cheung, Luo, Sia and Chen, 2009; Shu and Scott, 2014).
In this study, the two components of the IAM will be considered (see figure 2), argument quality and source credibility (Sussman and Siegal, 2003) and how these aspects of messages
affect people (Shen, Cheung, and Lee, 2013; Sussman and Siegal, 2003). Argument quality
focuses on the core of the message and takes in consideration that individuals carefully consider presented issues (Sussman and Siegal, 2003) and the persuasive strength of the arguments (Bhattacherjee and Sanford, 2006). Source credibility occurs when individuals instead of analyzing the content rather use simple decision rules to evaluate the message (Petty and Cacioppo, 1986; Sussman and Siegal, 2003). Source credibility is defined as the extent to which an information source is perceived to be believable, competent and trustworthy by the information recipients
Figure 2: Information Adoption Model Therefore, the following hypothesis are proposed:
H3: Argument quality on SNM sites positively affects millennials’ attitude towards information usefulness.
H4: Source credibility on SNM sites positively affects millennials’ attitude towards information usefulness.
2.2.4 Social Interaction
Social influence can be described as a change in an individual’s thoughts, opinions or actions based on interaction with other individuals or a group. This is often done in order to better fit into a group or situation by meeting the expectations of others. Sometimes, the individual may form the intention to perform a behavior even if the person is not favorable towards it (Chien, Kurnia and von Westarp, 2003). However, individuals believe that others expect them to perform that behavior, like visiting an intensively discussed restaurant on SNM. The impact can come from the external (e.g. mass media) or interpersonal (e.g. friends, family, colleagues) environment (Bhattacherjee, 2000).
Figure 3: Social Interaction Hence, it should be hypothesized:
H5: Social integration positively affects millennials’ subjective norm. H6: Social influence positively affects millennials’ subjective norm.
2.2.5 Perceived Risk
in making an overall evaluation of the sites and 3) the difference between the customers’ expectations and experiences in the service (Gunawan and Huarng, 2015). Risks regarding performance, financial, social and psychological will be covered in this study.
The last hypothesis is proposed as the following:
H7: Millennials’ perceived risk negatively affects the purchase intention of SNM marketed restaurants.
2.2.6 Summary
Figure 4: Conceptual Model (Gunawan and Huarng, 2015)
Table 1: Hypotheses
H1: Millennials’ attitude towards information usefulness positively affects the purchase intention of SNM marketed restaurants.
H2: Millennials’ subjective norm positively affects the purchase intention of SNM marketed restaurants.
H3: Argument quality on SNM sites positively affects millennials’ attitude towards information usefulness.
H4: Source credibility on SNM sites positively affects millennials’ attitude towards information usefulness.
H5: Social integration positively affects millennials’ subjective norm. H6: Social influence positively affects millennials’ subjective norm.
3 Method
3.1 Research Design
The purpose of this study was to analyze the effects of SNM on customers’ purchase intention. By investigating the relationship of the three determinants (attitude towards information usefulness, subjective norms and perceived risk) towards purchase intention, this study was by its nature explanatory since it allowed to establish casual relationships (Saunders, Lewis and Thornhill, 2007). An often-used concept connected with an explanatory design is the deductive approach as hypotheses often are derived from former research based on existing theory. A deductive approach allows suggesting possible reasons for particular relationships between variables (Saunders et al., 2007). By reviewing the existing research and theory in SNM, eWOM and purchase intention, hypotheses were formed. Here, a model developed by Gunawan and Huarng (2015), combining the Theory of Reasoned Action with the Information Adoption Model and perceived risk, was used for inspiration and adapted to the setting of this study. The model contained seven hypotheses that could be empirically tested through verification and falsification. To answer the research question, a quantitative research approach was adopted in order to investigate the hypothesized relationships between variables. Web-based questionnaires were distributed, which according to Saunders et al. (2007) is favorable for explanatory purposes where relationships between variables are to be determined. Apart from an increased reliability through a larger data set, this allows to enable testing and generalizing theory (Creswell, 2003; Saunders et al., 2007).
3.2 Research Strategy
questions in this study. Another advantage with questionnaires in quantitative explanatory studies is that a large amount of standardized data can in a highly economical way be gathered in a short period of time (Bryman and Bell, 2011; Saunders et al., 2007).
3.3 Sampling Frame
In order to investigate how eWOM in SNM impact the purchase intention of millennials in the restaurant setting, the target sample had to fulfill two criteria. The first criterion was a certain age-range, following the selected target group. As millennials are born between around 1980 and 2000 (Gurău, 2012; Tyler, 2007), they consequently had to be between 16 and 37 years old for this study. Active usage was the second criteria, where individuals needed to use SNM on a daily basis. It is argued that millennials embrace SNM (Moore, 2012) and that technology takes part in their everyday life (Prensky, 2001), leading toward the expectation of a high usage of SNM. A daily usage was considered to have an impact in forming their attitudes and beliefs and finally impacting their purchase intentions. As stated by Moore (2012) and Prensky (2001) millennials are considered as technological natives and active users embracing SNM in general, this study did not focus on specific SNM. If respondents did not fulfill these two criteria, they were disqualified from being considered for the analysis. The way the sample frame is defined leads to implications regarding the extent of generalizability (Saunders et al., 2007). Setting these two criteria in the study helped to improve the generalizability as it makes the sample as representative as possible for the studied population.
3.4 Measurements and Scales
reliability and validity was increased. Variables were transferred into closed-ended questions. The items were measured on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). Even though the research by Gunawan and Huarng (2015) used a 5-point Likert scale, a 7-point Likert scale was chosen, as people tend to avoid selecting the extremes at both ends of the scale (Saunders et al., 2007). By offering seven options, a more detailed range of answers was given.
Within this section, the eight measurements will be further addressed regarding their operationalization.
Argument quality
Argument quality is referred as the persuasive strength of the message (Bhattacherjee and Sanford, 2006). As customers use SNM to share experiences (Taylor and Todd, 1995) and enjoy discussing services on SNM sites (Hsu and Lin, 2008), it is in this study of importance to further examine the construct argument quality. Therefore, questions were asked about how people are engaged in reading eWOM in SNM as well as how information is perceived as sufficient for influencing the purchase intention.
Source credibility
Source credibility highlights that information can be perceived differently amongst customers regarding if the source is believable, competent and trustworthy (Sussman and Siegal, 2003). The expectation of this study is that credibility of information shared via SNM is related to the usefulness of eWOM information and purchase intention. This was tested with questions based on the previously mentioned nature of source credibility.
Attitude towards information usefulness
Social integration
The independent variable social integration refers to the social interaction of people with their strength and sincerity. The more social interaction takes place, the greater is the amount of the information exchange, leading towards an increasing influence on subjective norm (Chiu et al., 2006). In addition, customers tend to have greater trust in information that are believable and transparent (Cheung et al., 2008). Therefore, it is worth examining this construct in the context of eWOM in SNM with questions regarding the intensity of interaction and their accuracy. Social influence
The independent variable social influence argues that the usage of services by other people impacts the own attitudes and beliefs (Chien et al., 2003). As SNM focuses on the users existing network, only the interpersonal environment is considered. In conclusion, the item was measured by investigating the interpersonal impact on the subjective norm, assuming that if family members, friends or co-workers believed that eWOM about restaurants on SNM was useful, then the person might agree and finally be impacted in their purchase intention.
Subjective norm
Based on the TRA, subjective norm influences the intention to perform as it contains the degree to which a user perceives others’ expectations regarding a certain behavior (Hsu and Lin, 2008; Taylor and Todd, 1995). Hence, this item will be tested based on questions regarding the expectations of others towards the usage of SNM recommended restaurants.
Perceived risk
In this study perceived risk will be grouped into four different types of risk: performance risk (Lin, Lee and Horng, 2011), financial risk (Stone and Grønhaug, 1993), social risk (Corbitt, Thanasankit and Yi, 2003) and psychological risk (Corbitt et al., 2003), that customers can face with eWOM in SNM’s.
Purchase intention
(Taylor and Todd, 1995) and the prediction of this study is that millennials purchase intention is influenced by eWOM information.
The following table 2 summarizes the study’s construct measurements (questions in full can be found in appendix 1).
Table 2: Constructs and Items
Variable Question Reference
Argument quality AQ1, AQ2, AQ3, AQ7 AQ4, AQ5, AQ6
(Gunawan and Huarng, 2015) (Park et al., 2007)
Source credibility SC1, SC2
SC3, SC4, SC5, SC6
(Cheung et al., 2008)
(Gunawan and Huarng, 2015) Attitude towards information usefulness AT1, AT4 AT2, AT3 AT5, AT6
(Taylor and Todd, 1995) (Davis et al., 1989) (Hsu and Lin, 2008) Social integration SINT1, SINT2, SINT3,
SINT4, SINT5, SINT6, SINT7
(Chiu et al., 2006)
(Gunawan and Huarng, 2015)
Social influence SINF1, SINF2
SINF3, SINF4, SINF5
(Chien et al., 2003) (Bhattacherjee, 2000) Subjective norm SN1, SN2
SN3
(Hsu and Lin, 2008) (Taylor and Todd, 1995) Perceived risk PR1, PR2, PR3, PR4, PR5
PR6
(Corbitt et al., 2003)
(Stone and Grønhaug, 1993) Purchase intention PI1
PI2, PI3 PI4
(Gunawan and Huarng, 2015) (Taylor and Todd, 1995) (Hsu and Lin, 2008)
3.5 Research Procedure
3.5.1 Pilot-test
validity (Saunders et al., 2007). It is argued that the minimum number of participants in a pilot-test is ten respondents. Furthermore, it is important to take in consideration that respondents of the pilot-test include sufficient variations of the population (ibid). The pilot-test in this study included in total 20 respondents, a mix of millennials with different nationalities. As recommended by Saunders et al. (2007) questions were asked after respondents had submitted their questionnaire regarding the overall experience, including how long it took to complete the questionnaire, the clarity of instructions, if any questions were unclear, if it was easy to answer the questions and if they had any concerns about the layout. Minor semantic adjustments were made to make the questions comprehensible for the respondents. The final questionnaire can be found in appendix 1.
3.5.2 Data Collection
Questionnaires in this study were electronically distributed between the 24th of March and the
6th of April 2017. Google forms was the platform used for the questionnaire. The questionnaire
was distributed via Facebook, as this is the most common used SNM (Statista, 2016). The questionnaire was spread on personal profiles, in groups and by messages. A non-probability method with snowball sampling was used, as it allows easy access to a great sample at a low cost. This approach was encouraged as the contacted individuals were asked to forward the questionnaire to other possible respondents. The fulfillment of the criteria was furthermore certified, by having questions regarding age and SMN usage. Kosinski, Matz, Goling, Popov and Stillwell (2015) state several benefits, that Facebook offers as a tool for snowball sampling. The size and reach of Facebook allow to acquire a large and diverse sample size in the least expensive way. However, there is the possibility of sampling-selection biases, since users sharing the survey are likely to interact with others similar to themselves (Kosinski et al., 2015). That, in turn, could question how valid or representative the sample is (Bryman and Bell, 2011; Saunders et al., 2007). Considering the possible bias, this study had certain criteria that respondents needed to fulfill to increase the validity of this study.
In total 187 completed surveys were collected, and 163 surveys passed the criteria set for this study, born between 1980-2000 and using social media on a daily basis.
3.5.3 Data Analysis
cause distorted statistics (Hair, Black, Babin and Anderson, 2010).
When data was cleaned, the analysis started with a factor analysis to have a more manageable number of dimensions, since it summarizes a large set of variables down to a smaller number of dimensions (Pallant, 2011). Here, different steps were included: Kaiser-Meyer-Olkin and Bartlett’s test of sphericity measure the adequacy of the sample (ibid). Principal components analysis is a technique to determine “the smallest number of factors that can be used to best represent the interrelationships among the set of variables” (Pallant, 2011, p. 183). A Varimax rotation was used in this study as it presents patterns of loadings and which variables “clump together”, as well as it minimizes the number of variables for each factor based on high loadings. Afterwards, the reliability of the scale was assessed, which indicates how free it is from random error (ibid). Cronbach’s alpha coefficient was used as it according to Pallant (2011) is a common indicator for internal consistency.
As a next step in this study, three multiple regressions were conducted. According to Pallant (2011), this is an established tool to examine relationships between one continuous dependent variable and a number of independent variables or predictors, which was in line with the aim of this study. Before running the three multiple regressions, normality of data was checked as suggested by Pallant (2011). For the first two multiple regressions, the constructs of attitude towards information usefulness as well as subjective norm were the dependent variables. Within the third run attitude towards information usefulness and subjective norm together with the perceived risk formed the independent variables, where their impact toward the dependent variable of purchase intention was tested. By running a multiple regression, the hypotheses could be either falsified or verified.
3.5.4 Ethical Considerations
4 Results
4.1 Sample Characteristics and Data Cleaning
4.1.1 Sample Characteristics
As previously highlighted, 187 completed surveys were collected, 122 respondents were females and 65 respondents were male. In total, 163 surveys passed the criteria’s previously set for this study, millennials using social media daily. 24 nationalities were presented, where Germany and Sweden were most frequent. The most common accomplished degree was a graduate degree, with a percentage of 48%. Generalizability of this study’s findings was strengthened as similar characteristics were found. The current occupation of the respondents was as followed: 37% were working, 38% studied at a university and 17% were studying at a university and working part-time. In total had 86% of the respondents over 6 years of experience of using social media. Looking at the eating out habits, the majority with 52% of the respondents ate out 1-2 times a week, 15% ate out 3-4 times a week and 7% ate out more than 5 times a week. The remaining 26% ate out less than one time per week.
4.1.2 Cleaning Data
It was important to check for errors in the data set before analyzing data (Pallant, 2011; Saunders et al., 2007). As recommended by Pallant (2011) the first step was to create a codebook were all data was documented, defined, labeled and assigned numbers to each of the variables. The second step as suggested by Pallant (2011) was to check if values in the data set only range from what had been decided in the questionnaire. Data was checked for missing values but no missing values were found in this study. The distributed questionnaire only had mandatory questions and respondents could not continue to the next page before answering every question as a way to avoid missing data (ibid).
4.1.3 Outliers
After the data was cleaned, it was further examined with an outlier test. Outliers are extreme values, that can be either well below or well above in comparison with the rest of the data set (Hair et al., 2010; Pallant, 2011) and might cause non-normal data and distorted statistics (Hair et al., 2010). There are normal outliers and extreme outliers, depending on how far the data point are from the boxplot (Hair et al., 2010; Pallant, 2011).
distorting the statistics. Values for extreme outliers that only could be detected in one or two boxplots were changed towards a more neutral value on the Likert scale in accordance with Pallant (2011). However, one respondent was found to be an extreme outlier in every boxplot and this respondent was removed from the data set, ending with 162 approved respondents.
4.2 Construct Validity and Reliability
In order to assess construct validity between the eight variables, an exploratory factor analysis using principal component analysis with varimax rotation was performed. Furthermore, the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity were used to evaluate the intercorrelations between the variables (Kaiser, 1970). Since the sample size was 162 (after removing respondents that did not fulfill the prerequisites and the extreme outlier) it exceeded the 10 to 1 ratio recommended by Nunnally (1978), were ten cases for each item to be analyzed are advised. Other researchers see a minimum of 150 cases as sufficient (Pallant, 2011), which was also achieved in this study.
A first factor analysis was done, including all 36 items. However, an eight-factor solution describing the theoretically independent constructs could not be achieved. A step-by-step
process recommended by Hair et al. (2010)was adopted, that has similarly been used by Smith,
Milberg and Burke (1996) and Sethi and King (1994). Here, the factors were first evaluated in isolation, then in different pairs and finally as a collective network in connection with the conceptual model.
After running the analyses pairwise with each possible constellation and comparing them, first measurements were removed to get rid of the biggest disruptive measures. This was done if they cross loaded on the same factor or did not “clump together”. For example, did social integration (SINT) 1-4 load on a different factor than SINT 5-7. After comparing the content of the questions with the focus of the theoretical concept, SINT 5-7 were removed, as SINT 1-4 were considered to be more appropriate to reflect the concept of social integration. As factor analysis is a rather subject method, it is up to the researchers logical thinking to interpret the proceeded data (Pallant, 2011).
0,623, which is above the recommended minimum of 0,6 (Hair et al., 2010). Likewise, Bartlett’s test of sphericity was significant, since it was p=0,000 and therefore at the 1% level as recommended by Field (2000). Hence, it proved that the data set was appropriate for a factor analysis. Factor loadings below the value of 0,5 were excluded, because these loadings are considered to be poor and not significant enough (Hair et al., 2010). Moreover, cross-loading variables were removed to increase construct validity. The remaining factors with their loadings showed construct validity and are presented in table 3.
Table 3: Rotated Component Matrix - Factor Analysis 1
Rotated Component Matrix Component 1 2 AQ1 0,816 AQ2 0,604 AQ3 0,784 SC3 0,932 SC4 0,892
The next factor analysis evaluated the constructs of social integration (SINT) and social influence (SINF). Here, a two-factor solution was realized. The eigenvalues amounted 2,497 and 2,289, explaining 68,37% of the total variance, a KMO of 0,689 and a significant Bartlett’s test of sphericity with p=0,000 was achieved. As proceeded before, cross-loadings and low factor loadings were removed to achieve construct validity. The remaining measures with their loadings are presented in table 4.
Table 4: Rotated Component Matrix - Factor Analysis 2
Finally, a factor analysis was run with the constructs of attitude towards information usefulness (AT) subjective norm (SN), perceived risk (PR) and purchase intention (PI). Here, a four-factor solution was forced, with an eigenvalue of the fourth component of 0,839, which does not reach the original cut-off of 1. However, according to Pallant (2011), including a factor with an eigenvalue below 1 is an appropriate procedure, as it is up to the researcher to determine the number of factors that best describes the relationships among the variables. Two conflicting needs must be balanced: “the need to find a simple solution with as few factors as possible; and the need to explain as much of the variance in the original data set as possible” (Pallant, 2011, p. 183). In previous research (see table 2), the same questions have been used to measure the constructs, which in turn supports forcing a four-factor analysis, even if the fourth factor had an eigenvalue of 0,839. For this four-factor solution, the KMO was 0,857, Barlett’s Test of Sphericity significant (p=0,000) and the four factors described 75,76% of the total variance. The rotated component matrix for this factor analysis is presented in table 5.
Table 5: Rotated Component Matrix - Factor Analysis 3
Afterwards, internal reliability was tested using Cronbach’s alpha, since it is one of the most used techniques for this procedure, aiming to understand whether the measures in the same scale are consistent or not (Bryman and Bell, 2011). A recommendation is a Cronbach’s alpha coefficient value with a minimum of 0,7 (Nunnally, 1978). Looking at the results provided, reliability could be achieved for each measure, except for argument quality where the value was 0,574, which is considered as poor. Therefore, the mean inter-item correlation value was checked, which is recommended by Pallant (2011). Here, a value below the recommended range was also found, which pointed out the necessity to improve the items’ reliability. Looking into the correlation of the measurements, low values for AQ2 could be found. This item was therefore removed and Cronbach’s alpha was tested again, stating a value of 0,691. Since it is just slightly below 0,7, this value was accepted. The final Cronbach’s alpha values are stated in table 6. In sum, construct validity and reliability could be achieved for all eight items.
Table 6: Cronbach's Alpha
Measure Cronbach´s Alpha Number of Items
AQ 0,691 2 SC 0,827 2 AT 0,880 5 SINT 0,744 4 SINF 0,898 3 SN 0,945 3 PR 0,777 3 PI 0,814 3 4.3 Normality
center (Pallant, 2011). This is a minor violation in the study and data was still considered to be robust. If the distribution were perfectly normal, the values for skewness and kurtosis would be 0, however, this is an uncommon occurrence (Pallant, 2011). After normality check, the next step in this study was the hypotheses testing.
Table 7: Test of Normality - Skewness and Kurtosis Descriptive Statistics
N Min Max Mean Dev. Std. Skewness Kurtosis
Statistic Statistic Statistic Statistic Statistic Statistic Error Std. Statistic Error Std.
AQ 162 2 14 9,26 2,24 -0,527 0,191 0,275 0,379 SC 162 2 14 9,96 2,90 -0,999 0,191 0,715 0,379 AT 162 6 28 20,14 4,32 -0,959 0,191 1,181 0,379 SINT 162 6 28 20,16 4,55 -0,595 0,191 -0,041 0,379 SINF 162 3 21 12,10 5,22 -0,225 0,191 -0,854 0,379 SN 162 3 21 12,55 4,62 -0,429 0,191 -0,325 0,379 PR 162 3 21 10,44 3,64 0,310 0,191 -0,265 0,379 PI 162 3 21 12,96 3,98 -0,358 0,191 -0,190 0,379 4.4 Hypotheses Testing
To test the hypothesized relationships between the variables, three multiple regressions were conducted. This procedure allows assessing the relative impacts of independent variables on a dependent variable. However, different conditions must be fulfilled to do so.
First, it requires a sufficient sample size, where the required amount is taking the number of independent variables into account (Tabachnick and Fidell, 2013), stating N > 50 + 8m (m = number of independent variables). As this study has two or three independent variables, leading towards a required sample size of 66 and 74, it reaches this condition by having 162 respondents.
tolerance between 0,651 and 1 were found and therefore the values were well beneath the recommended number, which in turn demonstrated a low collinearity.
Third, as multiple regression analysis is sensitive to outliers, they must be removed. This had been done during the data cleaning.
Finally, data must be checked for normality, linearity, homoscedasticity and the independence of residuals, which refer to aspects of the distribution of the scores and the nature of the relationships among the variables (Hair et al., 2010). Here, Pallant (2011) suggests performing different plots where a reasonably linear pattern appeared. A rectangular, centralized pattern was found in the Partial Regression Plot for two of the regressions, suggesting no violations of homoscedasticity as well as the Normal Probability Plot confirmed normality, showing reasonably straight diagonal line from bottom left to top right. When testing the relationships among subjective norm as a dependent variable and social interaction and social influence as independent variables, one Partial Regression Plot did not show a rectangularly distributed plot, where the score are concentrated in the center. However, a clear or systematic pattern could not be found either. Neither did the normal probability plot show any violations of normality. In combination with the non-existing risk of multicollinearity that was assured before, it was decided to accept the results and see the assumptions as not violated.
Table 8: Multiple Regression IAM
B (unstandardized) Sig. (p-value)
Argument Quality (H3) 0,751 0,000
Source Credibility (H4) 0,523 0,000
R² 0,365
Model Sig 0,000
For the second multiple regression, subjective norm was the dependent variable and social influence and social integration the independent variables. Table 9 illustrates the outcomes, which could confirm H6. However, was H5 was not significant and therefore falsified. The confidence interval was kept at a 95% level, due to the p-value of 0,283, meaning a slight change of the interval could not have put H5 on a significant level.
Table 9: Multiple Regression Social Interaction
B (unstandardized) Sig. (p-value)
Social Integration (H5) 0,060 0,283
Social Influence (H6) 0,638 0,000
R² 0,522
Model Sig 0,000
Table 10: Multiple Regression TRA and Perceived Risk
B (unstandardized) Sig. (p-value) Attitude towards Information Usefulness (H1) 0,430 0,000
Subjective Norm (H2) 0,294 0,000
Perceived Risk (H7) 0,046 0,456
R² 0,515
Model Sig 0,000
Consequently, five out of seven hypotheses were verified, and the hypotheses rejected were social integration (H5) and perceived risk (H7). Figure 5 summarizes the relationships between the different variables, illustrating the impact of the different determinants of eWOM in SNM concerning the purchase intention of millennials in a restaurant setting.
5 Analysis
Hypothesis 1: Attitude towards information usefulness positively affects the purchase intention The first hypothesis (H1) of this study proposed, that attitude towards information usefulness positively affects the purchase intention of SNM marketed restaurants. H1 is accepted as there is a positive effect (B=0,430) of attitudes towards information usefulness on purchase intention. This goes hand in hand with previous literature (Ajzen and Fishbein, 1980) claiming that attitude is the main predictor of purchase intention. Millennials, that have a positive attitude towards the information read, learn a certain behavior (Wang et al., 2012) and are more likely to visit that restaurant. If they have a positive attitude towards reading posts they are likely to transfer the presented information to their own behavioral intention, meaning that they feel comfortable about the restaurant as well.
Hypothesis 2: Subjective norm positively affects the purchase intention
The second hypothesis (H2) proposed that subjective norm is positively influencing the purchase intention. The empirical results support this positive relationship (B=0,294), meaning that millennials take the recommendations of others into account and are impacted by their expectations regarding visiting a certain restaurant. This is in line with theory, which demonstrates a social pressure of what other people want, whether to perform or not perform a behavior (Ajzen and Fishbein, 1980). Hence, one can conclude, that if a restaurant was positively discussed on SNM, it leads towards other people recommending visiting a specific place which in turn puts pressure on millennials to finally behave as expected. Since millennials often seek for a group-belonging, visiting a certain restaurant can be a medium to express this belonging and their social status. Moreover, one can see that not only the SNM environment is impacting millennials, but also people from the direct social environment as family and friends, who take notice of restaurants discussed on SNM.
millennials consider the information to be useful, leading them to look more closely at the discussed restaurants. The core of the message is important as individuals consciously consider the given information (Sussman and Siegal, 2003). Considering that millennials are “digital natives” and use technology for sharing and creating content (Tapscott, 2009) it is of importance that the posts are of good quality even if posted from someone in their own network. For the persuasive strength (Bhattacherjee and Sanford, 2006) the content of the published post needs to be of good quality to impact millennials’ attitude towards the information.
Hypothesis 4: Source credibility positively affects attitudes towards information usefulness In this study was the fourth hypothesis (H4) that source credibility positively affects millennials’ attitude towards information usefulness. Empirical results (B=0,523) support that there is a direct positive effect of source credibility on attitudes towards information usefulness on SNM, meaning that H4 is accepted. If millennials consider the source to be credible, competent and trustworthy (Petty and Cacioppo, 1986), it creates a positive attitude. As this generation struggles with independently decision making (Howe and Strauss, 2007; Tyler, 2007) the source offers them security before deciding to visit a restaurant that has been discussed on SNM. Theory states, that individuals use simple decision rules instead of analyzing the content (Petty and Cacioppo, 1986; Sussman and Siegal, 2003), which is confirmed by the empirical data of this study.
Hypothesis 5: Social integration positively affects subjective norm
Hypothesis 6: Social influence positively affects subjective norm
Hypotheses 6, suggesting that the social influence is affecting the subjective norm could be supported by the empirical findings (B=0,638) which highlight a strong relationship between these two variables. One can conclude, that the expectations of the surrounding social environment to go to a restaurant that has been discussed on SNM lead towards a social pressure to visit that place. In line with the theory, a change in millennials’ behavior or their thoughts is the consequence of recommendations by their family or friends (Tyler, 2007). One reason for this behavior might be to better fit into a group as suggested by Tyler (2007). As millennials are described as a group that seeks advice by their close environment (Tyler, 2007), this results supports this behavior in the restaurant setting, illustrating the power of surround people on their decision making.
Hypothesis 7: Perceived risk negatively affects purchase intention
6 Discussion
Targeting millennials on SNM within the restaurant setting has already been highlighted as an interesting approach. Millennials are strongly engaged in SNM, have a high purchasing power and like visiting restaurants (Barton et al., 2012; Rohm et al., 2013; Schewe et al., 2013). Therefore, this generation appears to be a target group marketers can reach via eWOM shared among millennials’ networks on social media. By investigating different determinants, this study aimed to find out how different elements impact millennials purchase intentions within the service setting of restaurants.
In this study, the impact of three determinants towards purchase intention was tested: attitude towards information usefulness, subjective norm and perceived risk. Of these three determinants attitude towards information usefulness was found to be most relevant (B=0,430). This is in line with the TRA by Ajzen and Fishbein (1980), who argues for a strong connection between the two variables attitude and purchase intention. When millennials see posts about restaurants with a solid and convincing argumentation and perceive reading these posts as pleasant, their purchase intention will be impacted as they internalize the content of the post. Meaning, that attitudes are learned from these interactions. Therefore, millennials need to be convinced by the information usefulness in order to create a positive attitude towards their purchase intention. In addition to argument quality and source credibility, millennials’ interest in restaurants offers good support in creating a favorable attitude. This study’s empirical data confirmed that millennials have a positive opinion about positively discussed restaurants, which is supported by Millennial Marketing (2017) stating that they seek out friends’ opinions before deciding for a restaurant and are open to new advice. Simply said, this industry offers a good environment to benefit from the strong relationship between attitude towards information usefulness and purchase intention.
