Co-branding initiatives in relation to social media reputation:
The H&M yearly designer collaborations and the organizational
Instagram account.
Master’s Thesis – Master’s program Corporate Communication
Thesis supervisor: Piet Verhoeven
Papadogkona Ifigeneia
Student nr: 10841970
January 27
th, 2016
Abstract
This study sets out to examine whether there is a variance in social media reputation related to organizational co-branding initiatives. It reaches out to well-rounded corporate communication theories, such as agenda setting and framing, distinguishing two reputational dimensions: visibility and favorability. By adopting the understanding of framing as the collective process of meaning construction, taking advantage of the opportunities offered by social media metrics and adapting Deephouse’s coefficient of media favorability, an operationalization of social media reputation is developed. The hypotheses are tested employing a content analysis on H&M’s Instagram account. The results show a strong significant positive association between social media visibility and recency, suggesting that the number of likes increases strongly as time passes by. This is also the case when comparing between the three collaborations in question in terms of social media visibility, which could be explained by the overall increasing popularity of the medium, and as shown by the data, the organizational account, over time. In addition, the empirical analysis indicates that the two reputational dimensions, namely visibility and favorability, are negatively associated. No significant relation was established between social media reputation and co-branding, however, this study calls for the development of an extensive and sophisticated measurement tool in regard to social media reputation, which accounts for the content of the evaluations as well as the evaluative tonality (valence framing).
Introduction
Nowadays the participation of organizations on the Web is an indisputable fact. Organizations engage in the online environment to disseminate information, promote their products, build relationships with stakeholders but also to gather information, receive feedback and engage in a conversation with the public. Kietzmann, Hermkens, Mccarthy and Silvestre (2011) recognize reputation as one of the seven building blocks of social media for organizations. Corporate reputation, which has been defined as the overall evaluation of a firm by its stakeholders, produced by the interactions of the firm with its stakeholders, is considered a resource that leads to competitive advantage according to corporate communication research (Deephouse, 2000). Jones, Temperlay and Lima (2009), consequently argue that online reputation involves a corporate reputation created on the online environment and Kietzman et al. (2011) present how reputation on social media offers the functionality of monitoring the strength, passion, sentiment and reach of users and brands (Kietzman et al., 2011). In other words, social media, due to their structure, offer a direct form of evaluation on the organizational material posted, through ‘mechanical Turks’ (Kietzman et al., 2011) and by establishing new media metrics (Neill & Moody, 2014). Floreddu, Cabiddu and Evaristo (2014) further argue that the increasing use of social media determines that corporate reputation is influenced not by what firms do or say, but by how Internet users perceive firms’ actions.
This study focuses on the concept of social media reputation, derived from the broader concepts of corporate reputation and online reputation, as the corporate reputation ‘constructed’ in the online environment of social media and characterized by the environment’s specific types of metrics, through which users are enabled to evaluate the organization. The problem with media reputation research so far relies on the unidirectional creation of social meaning, in terms that meaning is created by the media. Thus, following Entman’s (2003) definition of framing, this study examines social media reputation in order
to account for the collectively constructed character of reputation that social media’s interactive structure allows for.
It is a common practice nowadays for organizations to collaborate and form marketing alliances (Ahn, Kim, & Forney, 2009). This popular technique, known as co-branding, is used by marketers in order to attempt to transfer the positive associations of the partner brands to a newly formed co-brand (Kahuni, Rowley, & Binsardi, 2009), therefore co-branding can be seen as a strategic alliance that connects two or more brands in the marketplace. Could this strategic tool affect an organization’s reputation by transferring the positive or negative evaluations of its stakeholders along as well?
Co-branding is an under-researched topic in corporate communication even though it has been vastly researched in marketing and business literature (Bucklin & Sengupta, 1993; Geylani, Inman, & Hofstede, 2008; Kippenberger, 2000; Rollet, Hoffmann, Coste-Manière & Panchout, 2013; Venkatesh, Mahajan & Muller, 2000; Wu & Chalip, 2013; Yi, Lee & Dubinsky,2010). Research examining co-branding, from a corporate branding perspective, in relation to corporate reputation, let alone social media reputation, has been scarce. Therefore this study will attempt to fill this gap and provide empirical evidence of co-brandings’ effects on social media reputation, if any, by examining the variation of an organization’s social media reputation before and after the collaboration. As for the practical implications, it will present communication professionals with a better understanding of social media reputation but also showcase whether co-branding can be employed by them as a tool to enhance their social media reputation.
Thus the research question states:
Theoretical background Corporate branding
In recent years, corporate brands have become valuable assets (Hatch, & Schultz, 2009a, 2009b). Corporate branding, revolves around the visibility of a company’s corporate brand in product communications and the relationship between corporate associations and product evaluations (Berens et al., 2015, as cited in Usunier, 2012). In other words, with corporate branding, instead of focusing on branding individual products, corporations promote their corporate brand, which, serving as a single ‘umbrella image’, incorporates a ‘panoply of products’ (Hatch, & Schultz, 2009a), bringing to marketing the ability to use the vision and culture of the company explicitly as part of its unique selling proposition (Hatch, & Schultz, 2003).
Hatch and Schultz (2003), highlight the interdisciplinary, in communication science, growing awareness that corporate brands can increase the company’s visibility, recognition and reputation in ways not fully appreciated by product branding. By presenting a comparison between the two, they demonstrate the shift in branding from the product to the corporation, which contributes not only to customer-based images of the organization, but to the images formed and held by all of its stakeholders. In addition, product branding has short term effects whereas corporate branding introduces a dynamic temporal dimension of heritage. They argue that, in comparison to product brand thinking, corporate branding puts stronger emphasis on the role of strategic vision as it requires top management’s reflections on who the company is and what it wants to become as well as the strategic coordination of all corporate communications. According to them, the successful corporate brand is formed by the interplay between the three foundational interconnected elements of corporate branding, which are strategic vision, organizational culture and the corporate images held by stakeholders. Strategic vision entails the ‘central idea behind the company that embodies and expresses top management’s aspiration for what the company will achieve in
the future’. Organizational culture refers to the internal values, beliefs and basic assumptions that embody the heritage of the company, while corporate images are defined as the views of the organization developed by its stakeholders.
Understandably, corporate branding has been treated as a way to influence corporate reputation (Schultz, Hatch, & Adams , 2012). Abratt and Kleyn (2012), argue that as stakeholders experience, relate to and commune with the brand, they are afforded opportunities to evaluate the brand on a number of core dimensions, which, when considered in totality over time, form the organization’s reputation. Schultz, Hatch and Adams(2012), more specifically, argue that corporate reputation captures the effects that brands and images have on the overall evaluations which stakeholders make of companies (van Riel & Fombrun, 2007: 40, as cited in Schultz, Hatch, & Adams , 2012 ). They further focus on a shift from regarding consumers as passive receivers of brand messages to multiple stakeholders as active co-creators of brand value, namely the co-creational perspective on corporate branding (Hatch, & Schultz, 2010). This constructionist approach focuses on corporate brands as vehicles for meaning that emerge from social interaction between the company and its environment, which is facilitated due to the Web 2.0 (Hatch, & Schultz, 2010; Melewar, Gotsi, & Andriopoulos,2012).
Co-Branding
The ‘brandization of society’ (Kornberger, 2010) refers to the increasingly important role of brands in the daily lives of consumers (Ambroise, Pantin-Sohier, Valette-Florence, & Albert, 2014). Brand managers are thus trying to maximize their potential impact through diverse branding strategies that aim at improving competitiveness and enhancing brand equity (Ahn, Kim, Forney, 2009). An increasingly popular and prevalent strategy among these is co-branding, a strategic alliance that connects two or more brands in the marketplace, strongly signaling to consumers the combined benefits of two quality brands together and making lasting impressions in the cluttered marketplace (Askegaard, &
Bengtsson, 2005; Geylani, Inman, & Hofstede, 2008). Kippenberger (2000) in his paper addresses co-branding as a ‘new competitive weapon’ and defines it as a distinctive and innovative strategy to address an increasingly competitive environment, that involves two or more brands with significant recognition, where all participating brand names are retained, that is of medium to long term duration. He further anticipates co-branding’s prevalence in the years to come. Indeed, firms are increasingly engaging in both intra-industrial collaborations as well as inter-industrial partnerships (Ahn, Kim, & Forney, 2010).
Rao and Ruekert (1994) argue that there are a variety of factors driving the surge in co-branding, ranging from the desire to gain access to new markets to the attempt to signal unobservable quality, and explain that co-branding follows the mathematical logic through which the combination of two brands creates not just the sum value of the two brands but an additional value based on the mutual strengthening of the two brand’s assets. In co-branding, each partner is assumed to have assets, capabilities or attributes that are complementary and constitute an alliance that makes conceptual sense to the consumer, with a ‘brand fit’ (Heslop, Nadeau, O'reilly, & Armenakyan, 2013). For example, each partner to a co-branding arrangement brings a customer base, which is potentially available to the other, thus expanding reach and awareness for both partner brands (Leuthesser, Kohli, & Suri, 2003).
The fashion industry has vastly employed co-branding, often entitled as ‘collaboration’, as a means to building differentiation and reputation (Rollet, Hoffmann, Coste-Manière, & Panchout, 2013; Wu, & Chalip, 2013). One well known example is the creative collaboration between fashion designers and mass-market retailers, such as the H&M yearly collaborations with high fashion houses (Rollet et al., 2013). As Rollet et al. (2013) argue, such co-branding initiatives transfer the attributes of uniqueness and rarity that high fashion and the luxury industry entails to the fast fashion organization. Kim et al.
(2014), on the other hand, focus on the awareness of consumers raised by collaborations, for both partner brands, while enhancing the fast-fashion brand’s image.
Considered from a corporate branding perspective, however, co-branding is not simply about a collaborative attempt in product branding. It is not about promoting a product, but instead, about promoting the co-brand, constituted by and representative of the two collaborating corporate brands. In other words, from a corporate branding perspective, co-branding is not about the product per se and its attributes, but about the two different organizational cultures collaborating together under a common vision, interlinking their corporate images in the eyes of their stakeholders. Co-branding literature has focused on brand fit, the logical fit between the images of both brands, as the most important success factor of the strategy (Kippenberger, 2000). In addition, research (Baumgarth, 2004; Kahuni, Rowley, & Binsardi, 2009; Keller & Lehmann, 2006; Leuthesser, Kohli, & Suri, 2003; Simonin and Ruth 1998; Washburn, Till, & Priluck, 2000) has shown that evaluations of the co-branded product may have an impact on consumers' perceptions of the partner brands (Simonin and Ruth 1998), namely spillover effects. These studies show that by engaging in a collaboration, the two corporate brands can affect each other’s image through spillover effects. However, corporate reputation research has given limited attention to reputational transfers under conditions of co-branding (Heslop et al., 2013), even though corporate branding has been linked to corporate reputation (Abbrat & Kleyn, 2012; Schultz, Hatch, & Adams, 2012).
Corporate reputation
Corporate reputation has been considered a strategic resource that can have a significant impact on business performance and lead to competitive advantage of a firm in corporate communication research (Abratt & Kleyn, 2012; Deephouse, 2000; Ewing, Caruana, & Loy, 1999; Fombrun, Gardberg, & Sever, 2000). According to Ewing, Caruana and Loy (1999), organizations with good reputations attract positive stakeholder
engagement while Fombrun et al. (2000) argue that reputation is rooted in the aggregated perceptions of the organization’s stakeholders.
Abratt and Kleyn (2012), in order to define corporate reputation, employ Gotsi and Wilson’s definition, as ‘A corporate reputation is a stakeholder’s overall evaluation of a company over time’ (Gotsi & Wilson, 2001, p. 29 as cited in Abratt & Kleyn, 2012). The overall evaluative character of reputation as a construct is central in the definition. Publics construct reputations from available information about a firm’s activities, originating from the firms themselves, the media or from other monitors (Fombrun & Shanley, 1990). Deephouse (2000), focusing on the information provided by the media, develops a variant of corporate reputation, entitled ‘media reputation’ and defined as the ‘overall evaluation of a firm presented in the media resulting from the stream of media stories about the firm’. This approach to corporate reputation can be explained by resorting to well established media effects theories such as agenda setting (Carroll & McCombs, 2003; Carroll, 2004; Kiousis et al., 2007; Meijer & Kleinnijenhuis, 2006) and framing (Carroll, 2004). Communication scholars have applied those concepts extensively in order to explore the relationship between media coverage and corporate reputation.
Corporate reputation and agenda setting
Carroll and McCombs (2003) address how a corporation’s exposure to news coverage can influence public opinion toward the corporation. Further empirical studies have demonstrated that media attention (how many news reports cover a corporation) and media favorability (how the news media portray a corporation) influence corporate reputation (Deephouse 2000; Kiousis et al. 2007; Meijer and Kleinnijenhuis 2006).
Agenda-setting theory implies that the salience of elements on the news agenda influences their salience on the public agenda (Carroll & McCombs, 2003). Lee and Carroll (2010), investigating media salience, such as the attention and prominence of an issue in the media, draw from Kiousis (2004, as cited in Lee & Carroll, 2010) who distinguishes
three dimensions: attention, prominence, and valence. Media attention refers to the media awareness of an object, while prominence emphasizes the relative importance of an issue. The third dimension, media valence, is the affective aspect of an object in the news and can be represented as the ‘tone toward the object of a story’.
First level agenda setting thus focuses on the effect of media visibility and assumes that the attention and prominence of given objects in news coverage has a direct effect on the salience of those objects within public awareness or discussion (Carroll, 2004). Based on first level agenda setting, Carroll (2004), found a correlation between media visibility and public awareness, thus linking visibility to corporate reputation.
At the second level, agenda setting theory argues that the salience of attributes on the media agenda influences the salience of those attributes on the public agenda, and it can be described in terms of two dimensions: substantive (cognitive) and evaluative (affective) (Carroll & McCombs, 2003). Deephouse (2000) refers to this evaluative dimension of the media’s second level agenda setting when introducing media favorability. He suggests that media not only convey information, but also make and present reputational assessments to their audiences. He employs valence framing, categorizing media pieces as favorable, unfavorable or neutral, in order to determine a firm’s overall media favorability, which he uses as an indicator of media reputation. Thus, the affective dimension of the second level agenda setting, as media favorability is also linked to reputation.
Consequently, media visibility and media favorability are the two factors examined by the different levels of the agenda setting theory as reputational dimensions.
Corporate reputation and framing
Second level agenda setting, also known as attribute agenda setting (Tong, 2013), suggests that the public will dynamically form their perceptions of the organization’s reputation based on what the media say (Carroll & McCombs, 2003; Deephouse, 2000). In
corporate communication literature, such information on attributes has been discussed from the framing perspective (Deephouse, 2000; Entman, 2003; Hallahan, 1999).
As a property of a message, a frame limits or defines the message's meaning by shaping the inferences that individuals make about the message (Hallahan, 1999). Framing is considered a critical activity in the construction of social reality since it helps shape the perspectives through which people see the world and has been treated as a collective process of meaning construction (Entman, 2003; Hallahan, 1999; Meer, Verhoeven, Beentjes, & Vliegenthart, 2014). Entman (2003) defines framing as ‘selecting and highlighting some facets of events or issues, and making connections among them so as to promote a particular interpretation, evaluation, and/or solution’.
In relation to media reputation, literature (Carroll & McCombs, 2003; Deephouse, 2000; Tong, 2013) has focused on valence framing. Valence refers to the embedded positive or negative sentiment, evaluation, or attitude toward the product or brand, which can be shown through the use of positive or negative words (Goh, Heng, & Lin, 2013). In other words, valence framing examines evaluative tonality, in terms of positive, negative and neutral. Deephouse (2000), more specifically, employing media reputation and media favorability interchangeably, measures the tonality of news coverage, classifying them as favorable, unfavorable or neutral, and measures the overall media reputation of an organization using the coefficient of media favorability, which will be presented in the methods section of this paper.
Levin, Schneider and Gaeth (1998), in their paper present attribute framing, as one of the three types of valence framing, which affects the evaluation of object or event characteristics. On this basis, they suggest that positive framing supports more favorable evaluations and that negative framing supports less favorable evaluations, based on the argument that attribute framing effects occur because information is encoded relative to its descriptive valence. Furthermore, Carroll and McCombs (2003), argue that ‘the more
positive media coverage is for a particular attribute, the more positively will members of the public perceive that attribute’. Thus, Tong (2013), suggests that reputation attributes in news coverage can be analyzed by the tonality of media favorability (Deephouse, 2000). Social media reputation
In his study, Luca (2011) suggests that online consumer reviews substitute for more traditional forms of reputation. Web 2.0, along with social media, a group of Internet-based applications that build on its ideological and technological foundations, allow for the creation and exchange of User Generated Content (UGC) (Kaplan & Haenlein, 2010) and empower eWordOfMouth (eWom). As Constantinides and Fountain (2008) put it, ‘the user is a vital factor for all categories of Web 2.0 applications, not only as a consumer but mainly as a content contributor’. Misopoulos, Mitic, Kapoulas, and Karapiperis (2014), in addition, argue that consumer commentaries on social media are the most sought and high valued, amongst online audiences, types of information online, creating thus a source of customer feedback ‘by the users for the users’. Thus, the internet, and social media in particular, have been theorized as being able to break the press’s monopoly on agenda setting, by making everyone equal and giving them a platform to express themselves when it comes to public debate (Jacobson, 2013).
According to Meriläinen and Vos (2011), the fact that Internet users are able to discuss issues online subsequently influences agenda setting. Groshek and Groshek (2013), also notice that this shift towards a converged media environment where the audiences are simultaneously media users and producers is transforming agenda setting. They argue that agenda setting is no longer a top-down controlled process from media to audiences, but also a dynamic process where online users are able, due to online spaces, to shape the public agenda. Jacobson (2013) also refers to a kind of Internet-enabled ‘‘metajournalism,’’ where the news audience may actively shape the news agenda by commenting on, sharing and rating the news.
Furthermore, if one conceptualizes framing as the collective process of meaning construction (Entman, 2003), social media, through UGC and eWOM, consist of a representative platform when it comes to the overall evaluation of an organization. Goh, Heng and Lin (2013), employing valence framing, demonstrate how valence, embedded in UGC, can be interpreted as the evaluations of a brand or product over the years while, at the same time, can surpass marketing generated content valence in driving purchases. According to them, the reason for that is that consumers have developed a tendency to be skeptical toward marketing messages, while being trustworthier of UGC. In their research, valence refers to the embedded positive or negative sentiment, evaluation, or attitude toward the product or brand, which can be shown through the use of positive or negative words. In other words, the more positive UGC is associated with a brand, the more positive the overall evaluation of the brand will be.
In this spirit, Floreddu, Cabiddu and Evaristo (2014) suggest that due to the increasing use of social media, corporate reputation is influenced not by what organizations do or say, but by how Internet users perceive organizations’ actions. Thus, instead of reputation belonging to the organization itself, it is to a large extent controlled and distributed by the organization’s stakeholders (Aula, 2011). After all, reputations are created through social realities that are formed narratively, as a collection of stories told about an organization, and these stories are more prolific and heterogeneous than ever on the Internet (Aula, 2011). In other words, through commenting on social media, the diverse online public can directly evaluate organizational actions and contribute to the collective formation of the organizational reputation online.
Consequently social media reputation could be defined as the corporate reputation ‘constructed’ in the online environment of social media and characterized by the environment’s specific types of metrics, through which users are enabled to evaluate the organization. Kietzmann et al. (2011), argue that reputation can have different meanings on
social media platforms and that each organization should identify the appropriate reputation metrics for its community’s social media engagement. Among these, there are objective data that can be gathered, such as number of views or followers, or data on the collective intelligence of the crowd. On a similar note, Neill and Moody (2014), present how brand culture is authenticated by the masses (Fournier & Avery, 2011), with social media expanding the industry’s metrics and allowing the audiences to directly evaluate their initiatives by 'liking', 'following', 'retweeting' or 'sharing'.
According to a report by Statista (2015), among the most popular, measured by active users, social media are Facebook, Twitter and Instagram. Facebook, which is well suited for rich content accompanied by images or videos, presents countless opportunities for engagement and has the largest percentage of active users, making it highly brand saturated (Bennet, 2014; Yu, 2014). Twitter, is a microblogging site, excellent for real time engagement with consumers, allowing information to be shared instantly while offering thought-provoking conversations. Instagram is a highly visual medium that has had an enormous amount of growth in a short amount of time (Vaynerchuck, 2015; Wagner, 2015), attracting 71% of the world’s largest brands. In fact, data from the Pew Research Center (2015) found that Instagram was the fastest growing major social network among U.S. adults last year while a report by Digital Net Agency (2012) suggests an increase in engagement with brands on Instagram, surpassing that of Twitter (Laird, 2013). Specifically for beauty and fashion brands, engagement and interactions on Instagram are higher than any other platform, while frequency of posting is gradually increasing (L2, 2015; Pathak, 2015). Instagram, which was acquired by Facebook in 2012 for $1 billion, is regarded as the most important social media platform for many fashion brands (Sherman, 2015), probably due to its power of visual storytelling, which is a good fit for the fashion industry.
Presenting the Hypotheses
In conclusion, drawing from the literature presented above, social media reputation can be treated using visibility and favorability as its two dimensions. In addition, the medium’s specific metrics alongside with UGC valence framing can be employed in its operationalization.
The research question examines whether co-branding has an effect on social media reputation. Co- branding is considered a popular marketing practice that has been used to transfer evaluations of co-brands into one product, while at the same time allowing for spillover effects between partner brands. In the fashion industry, it is suggested that the fast fashion brand is the one that will be affected more positively by the collaboration, increasing its reach to high fashion consumers and ‘adopting’ high fashion attributes such as rarity and uniqueness. Thus, based on the theoretical framework, it is argued that co-branding will increase the fast fashion brand’s social media reputation. As presented above, social media reputation involves two dimensions, namely visibility and favorability. Consequently the first hypotheses state:
H1a: Organizational posts regarding co-branding will score higher on social media visibility than regular organization posts.
H1b: Organizational posts regarding co-branding will score higher on social media favorability than regular organization posts.
However, considering reputation concerns the overall evaluation of an organization over time, it is expected that co-branding’s effect on social media reputation will also affect the evaluation of the organizational online activities for the duration of the collaboration. Thus, the long term effect of co-branding on social media reputation is addressed by the second pair of hypotheses:
H2a: Posts published after the announcement of the collaboration will score higher on social media visibility than posts published before the announcement.
H2b: Posts published after the announcement of the collaboration will score higher on social media favorability than posts published before the announcement.
Finally, considering that there are many cases where brands engage in co-branding with different partner brands on a yearly basis, it is expected that the effects on social media reputation will have an additive character. In addition, based on corporate reputation literature (Tong, 2013), which suggests that recency or immediacy of media coverage has a significant relationship with reputation regarding the direction and the total movement of the changes in organizational reputation, it is hypothesized that the more recent the collaboration, the higher the social media reputation. The last pair of hypotheses will thus allow to examine whether the different collaborations of the organization had a different impact on its social media reputation.
H3a: The yearly brand collaborations will differ significantly in terms of social media visibility, with the most recent one, scoring the highest and the least recent scoring the lowest.
H3b: The yearly brand collaborations will differ significantly in terms of social media favorability, with the most recent one, scoring the highest and the least recent scoring the lowest.
Methods The case of H&M
The focal organization chosen for this study is H&M (Hennes & Mauritz). H&M is a consumer goods, fast fashion organization, which has been a leader in fashion collaborations since 2004 (Rollet et al. 2014). For H&M, its co-branding strategy is a way of reinforcing its image, enhancing its awareness, generating extra income and opening its activity to new market segments according to Rollet et al. (2014). They define H&M’s practice as a form of ‘masstige’, a term used particularly in fast fashion to describe luxury or premium products at price points that fill the gap between mass and high-end. Thus, on one
hand, H&M, as a brand, achieves high visibility in a saturated segment and underpins its mission of offering customers fashion and quality at the best prices, while, on the other, the designer brand succeeds in maintaining high media presence and desirability without tainting the core values of its premium lines.
This research will focus on the three more recent H&M collaborations, with high fashion brands Isabel Marant (2013), Alexander Wang (2014) and Balmain (2015). Those yearly collaborations make it relevant to the research question, addressing the co-branding variable, but also allow for a deeper understanding of co-branding’s long term effects on reputation.
Design
In order to answer the research question, a quantitative content analysis was conducted on H&M’s Instagram account. Content analysis is a transparent, unobtrusive method that makes replication easier, while focusing on a single case allows obtaining extensive and detailed information on that case (Bryman, 2012).
Since social media reputation is to be measured, a content analysis on the company’s official Instagram account is conducted. According to eBizMBA (2015) Instagram is among the ten most popular social media, with the most engaged users (fastcompany.com, 2014) and many companies, acknowledging its popularity, engage in 'Instagramming' in order to share organizational news, promote their products but also to approach and maintain a relationship with consumers. Instagram, as a social medium, allows to visually present corporate projects and products for public evaluation, while at the same time, due to its structure, allows for a clear distinction between branded and co-branded projects. In addition, corporate communication research on Instagram is scarce, even though it is among the most popular social media and is vastly used for organizational purposes.
Instagram allows for an organization to have its organizational account where it posts material, namely the ‘post’, and other Instagram users are able to like or comment on that post. Each post, which can be either an image or a video and is presented with a title or description, namely the ‘caption’. Instagram users who choose to follow an account, thus making its posts directly visible in their timeline, are called followers. Users can also be directed to a post via Instagram search, for the account name or a specific hashtag, or via ‘mentions’, when someone tags them in a comment of that post. H&M’s Instagram account provides a good case for analysis, considering it is popular (10.6 million followers) and active (1,788 posts) (November 10th, 2015).
Sample
The sample for this research comprised of Instagram posts on H&M’s official Instagram account in the past three years, starting from January 2013 until the end of data collection. Alas, due to time limitations, it was not possible to investigate the entirety of posts for that period, therefore, every second post was included in the dataset, resulting in a systematic sample. A systematic sample was chosen instead of a random sample, in order to evenly include data from the desirable timeframe, considering the publication date plays a significant role in the analysis. Finally, not all posts from year 2015 were included in the sample, considering November 11th, 2015 was the end period for data gathering. Thus, it is a systematic sample in terms of organizational posts for the designated period, focusing on the last three yearly H&M collaborations; with Isabel Marant (2013), Alexander Wang (2014) and Balmain (2015). Both photos and videos were included in the sample and the sample unit is the Instagram post. This sampling procedure resulted in a sample of 803 posts.
For each post, its latest 150 comments were also coded. Alas, considering the number of comments on an H&M post ranges around a few hundreds, it was not feasible, due to time restrictions, to code the entirety of comments for all posts included in the
sample. Additionally, 150 comments per post is the number that Instagram proposes as acceptable for download, therefore, for practical reasons, this amount was chosen. However, this number varied in terms that not all posts exceeded the amount of 150 comments.
Procedure
The sample, consisting of H&M Instagram posts and their latest 150 comments, was gathered by executing a query for the aforementioned timeframe on Instagram’s developers’ API. The script employed (Appendix A) was written in Python. The data were downloaded simultaneously on November 11th, 2015, where each post and its relevant information appeared in a separate .csv file. More specifically, each file included information such as the name of the account, the number of followers, the number of posts, the unique identification of that post, the type of post, the timestamp, the caption and the number of comments on the post. For each post, a link to the original Instagram page, where the photo or video can be viewed, was included in the .csv file, in case there was a need for clarification. An example of such a file can be seen in appendix B. This information was to be transferred in the SPSS coding sheet, based on the coding instructions. The coding, which lasted three weeks (54hrs), was conducted based on the codebook provided (Appendix C). Ten percent of the data (n=80) were randomly selected, using an online random generator, and coded by a second coder, in order to calculate the inter coder reliability. Krippendorff’s Alpha as well as the coder percentage agreement, solely for categorical variables, were used to test the intercoder reliability. Even though the research unit is the Instagram post, the comment-focused intercoder reliability was also calculated for reasons of thoroughness. The results yielded high intercoder reliability ranging from the lowest value at Krippendorff’s A = 0.95 (variable Unfavorable) and the highest value at Krippendorff’s A = 1 (variable Relevance). The exact reliability values for all relevant
variables as well as the intercoder reliability procedure are extensively presented in appendix D.
Coders were instructed to explicitly follow the codebook’s instructions while coding. In addition, at the end of the codebook, a coding example was provided in order to orientate the coder. An SPSS file including all the relevant variables was used as the coding sheet, where each row represented an organizational post. In order to calculate the total amount of favorable, unfavorable, neutral and irrelevant comments for that post, a tonality sub-coding was additionally conducted on the .csv file. Using Excel’s 'sum' function, these values were calculated for each post separately and inserted in the coding sheet.
After all data were downloaded as .csv files and coded into the SPSS coding sheet, the dataset was ready for the analysis, which was conducted in SPSS.
Measures
Independent Variables.
Co-branding.
Variable Co-brandings is binary and shows whether the post is about one of the three collaborations in question. The coder should read the caption of the photo, which is entitled 'Caption' in the downloaded dataset, and in case there is a mention to Isabel Marant, Alexander Wang or Balmain, or if one of the hashtags #HMxIsabelMarant, #HMxAlexanderWang or #HMBalmaination is used he should code it as 1 (yes), otherwise he should code it as 0 (no).
Co_Brand.
Variable Co_Brand is a categorical variable, which helps distinguish in which of the three collaborations the post is referring to. This variable, is indicative of the collaborating brand thus should be coded only if variable co-branding is coded as 1, otherwise the coder should skip it. If Isabel Marant is mentioned in the caption or the #HMxIsabelMarant is used, it should be coded as 1, if Alexander Wang is mentioned in the caption or the
#HMxAlexanderWang is used it should be coded as 2 and if Balmain is mentioned in the caption or the #HMBalmaination is used it should be coded as 3.
Dependent Variables.
In order to operationalize social media reputation, drawing from agenda setting theory, social media visibility and social media favorability are used as indicators.
Visibility.
Variable Visibility refers to social media visibility. Drawing from Carroll (2004), who defines media visibility as ‘the aggregated news report relating to a specific company within a prescribed period’, and Lee & Carroll (2010), who link visibility to attention, social media visibility in this paper refers to the number of people who viewed an organizational post. Since Instagram does not allow to gather information on the variance of followers per time, making visible only the total number of followers in the present, the number of likes per post will be used as an indicator of visibility, a ratio variable.
Favorability.
Variable ‘Favorability’ is indicative of the post’s social media favorability that refers to the overall evaluative tonality for that post. In order to calculate social media favorability an adjusted version of Deephouse’s coefficient of media favorableness (2000) is used, based on the coding of the 150 most recent comments of each post. In order to do this, the coder should first code the comments of each post in the csv file. Each comment, which can be perceived as the recording unit, is rated as favorable, unfavorable or neutral following common practice in media research (Deephouse, 2000; Lee & Carroll, 2010). A summary of each rating follows.
Favorable comments are comments that are positive towards the organization, refer to it in a positive emotional appeal, as an object of respect and admiration, or particularly trustworthy. These may include comments that show support for the organization, that employ emoticons of appraisal (such as thumbs up, or hearts), that express anticipation for
an announcement, excitement about news or intent to buy the product. In addition, comments that directly address the organization in order to gain more information on its activities are included in this category, as well as comments from the organization responding to a request, since users will evaluate positively the organization in terms of being responsive and interactive (Fournier & Avery, 2011). Unfavorable comments refer to content that is unfavorable towards the organization, generating negative emotional appeal, or is portrayed as unworthy of admiration, respect, or trust. Such comments are those that badmouth the organization, express disappointment or frustration about the organization, disapproval of an announcement or employ angry emoticons. The essence of a neutral rating was the commentary without evaluative modifiers, having the absence of both positive and negative content. For example, a comment is rated as neutral when it simply tags or mentions someone, without providing further commentary or when it provides more information on a post, such as who the photographer is, without an evaluative tonality. This rating was also given when the tonality of the comment was mixed, balancing favorable and unfavorable (Deephouse, 2000). Comments unfit for any of the aforementioned categories are classified as irrelevant. Irrelevant comments are those that are not written in English, and therefore their content cannot be analyzed, promotional, spam or comments that have no association with the posted material nor the organization. For example, if someone wants to know when a product will be available in their country, even though the post is about an organizational event, the comment is coded as favorable, because it shows interest and anticipation towards the organization. If the post is about the model of the new H&M campaign and the users are arguing on whether or not it was a good choice, those comments are relevant and should be categorized accordingly. However, if the discussion reaches the point where users are arguing on something general, such as the standards of contemporary beauty, those comments are coded as irrelevant, as they no longer evaluate the organizational initiative.
The category present is marked as 1, while the rest are coded as 0. Afterwards, the total number of comments in each category is calculated and inserted in the coding sheet under the variable names Favorable, Unfavorable, Neutral and Irrelevant respectively. Those ratio variables indicate the total number of comments belonging to each of the tonality categories per post, and will be used in order to calculate social media favorability. Drawing from Deephouse’s (2000) coefficient of media favorableness operationalization and coding, it is possible to calculate the coefficient of social media favorableness for each Instagram post. Based on Deephouse (2000), its formula is:
(f2 – fu)/(total)2 if f > u; Coefficient of social media favorableness = 0 if f = u;
(fu – u2)/(total)2 if f < u;
where f = number of favorable comments in a post, u = number of unfavorable comments in a post; and total = the total number of comments coded for that post excluding the irrelevant ones, which in this case would be the sum of favorable, neutral and unfavorable comments. Favorability is thus a ratio variable, whose range is (- 1,1), where 1 indicates all positive comments, -1 indicates all unfavorable comments and 0 indicates a balance between the two in that single post.
Descriptive variables.
Some general descriptive variables that will facilitate the navigation through the sample, replication of the research as well as allow to draw general inferences are also included in the coding. Such variables are the name of the organization’s account (Account), a categorical variable, as welll as the number of followers (Followers) and the number of posts (Posts), which are ratio variables. This information will appear in each csv file next to the titles 'Organization', 'Followers' and 'Posts' respectively. This information is descriptive of the account and will not vary per post, considering this research focuses on a single account.
The next variable to be coded, Post_ID, a categorical variable, is the unique identification number of each post, as provided by the Instagram’s API. This number is indicative of the post being coded, thus allowing to navigate through the sample since it is also the name of the downloaded csv file.
Variable ‘Media’ is also categorical and will help identify whether the material posted is a photo, coded as 1, or a video, coded as 2.
Variable ‘Date’, is an ordinal variable indicative of the date the post was published. Even though it is classified in the descriptive variables, Date will be recoded accordingly during the analysis in order to divide organizational posts into groups, before and after the announcement of the collaboration per year, thus serving as a grouping variable.
Relevance, a binary variable, demonstrates whether the post is directly concerned with an organizational activity or initiative (coded as 1) or not (coded as 0). H&M’s Instagram, as most organizational accounts, includes some editorial posts that are not directly linked to the organization, such as promoting an artist’s new album or wishing happy birthday to a celebrity. Such posts, were marked as irrelevant and excluded from the analysis. Considering their evaluation is not directly connected with the organization, there is no need to code the tonality of the comments for these posts.
Finally, variable ‘Comments’ is indicative of the total number of comments per post. ‘Comments’ is a ratio variable, which will allow to draw general inferences from the data, however, is not directly used in order to test the hypotheses, considering that only 150 comments per post are employed to calculate the social media favorability.
Preparation of the dataset and analyses used
A single post, referring to the designer collaboration of a past year with MMM (Maison Martin Margiela), was declared missing. A frequency analysis was then run on all variables, in order to check their distribution. Even though ratio variables Visibility, Comments, Favorable, Unfavorable, Neutral and Irrelevant are not normally distributed, the central limit
theorem applies and normal distribution is assumed because of the large sample (n = 803). Finally, due to the structure of the dataset, there is no need to select cases during the analyses, since for irrelevant organizational posts the relevant variables have been left blank and thus assumed missing.
In order to test the hypotheses, T-tests and ANOVAs will be conducted in SPSS since different groups need to be compared. Even though effect sizes are usually omitted in content analysis studies, considering it is difficult to establish a direction in the relationship between two variables, in this study effect sizes are reported considering a direction is assumed, meaning that an independent variable, namely co-branding, is assumed to have an effect on a dependent variable, namely social media reputation (visibility and favorability), as stated in the research question. This choice is in line with the Graduate’s School of Communication (2012) reporting guidelines, and the lead from literature (Paul & Plucker, 2004; Plucker, 1997; Sullivan & Feinn, 2012), highlighting the importance of reporting effect sizes in order to understand, interpret, compare or even replicate findings.
Calculating the coefficient of social media favorability.
Before proceeding with the analysis of the data, dependent variable Favorability needs to be computed based on the sums of the valence framing on the comments, following Deephouse’s (2000) formula for the coefficient of media favorability:
(f2 – fu)/(total)2 if f > u; Coefficient of social media favorableness = 0 if f = u; (fu – u2)/(total)2 if f < u;
where f = Favorable, u = Unfavorable; and total = Favorable + Unfavorable + Neutral. Some studies do not include the number of Neutral comments in order to calculate the total amount of comments, however following Deephouse’s (2000) suggestion, Neutral comments are included in this calculation in order to obtain more reliable and representative results regarding favorability. For example assume that a post has no
unfavorable comments, 99 neutral comments and 1 favorable comment. In case neutral comments were not taken into consideration Favorability = 1, which is the highest value possible for the variable. However, when neutral comments are included Favorability = 0.01, a value more representative of the evaluation of the post and more representative than the former, since it accounts for a greater number of posts. Irrelevant posts are not included in the calculation of Favorability.
The formula assumes three conditions, if f > u, if f = u and if f < u, and uses f2, fu, u2 and (total)2. These values were calculated into new variables using the SPSS Compute function where f2 = Favorable * Favorable, u2 = Unfavorable * Unfavorable, fu = Favorable * Unfavorable and (total)2=(Favorable + Unfavorable + Neutral) * (Favorable + Unfavorable + Neutral). Having computed these variables, it is now possible to apply the formula, by using the function Recode Into Different variable, namely Favorability, minding the three conditions. Variable Favorability (-1, 1) is thus created. The syntax for this procedure can be found in Appendix E.1. Even though a frequency analysis on the variable shows that it is not normally distributed, the central limit theorem applies because of the large sample size (n = 663) and normal distribution can be assumed.
Defining the periods before and after each year’s designer collaboration.
Finally, variable Date, needs to be recoded into an ordinal variable that will facilitate the analysis. First, Date is recoded into a different variable, namely Year, where all posts from 2013 are coded as 1, those from 2014 as 2 and those from 2015 as 3. Then, based on the announcement dates of each yearly collaboration (Isabel Marant on 11.6.2013, Alexander Wang on 13.4.14, Balmain on 18.5.15) each year is split into two periods, one before and one after the collaboration announcement (including the date of the announcement). The end variable, namely CDateA is an ordinal variable where 0 = posts from 2013, before the Isabel Marant collaboration announcement, 1 = posts from 2013, after the Isabel Marant collaboration announcement, 2 = posts from 2014, before the
Alexander Wang collaboration announcement, 3 = posts from 2014, after the Alexander Wang collaboration announcement, 4 = posts from 2015, before the Balmain collaboration announcement and 5 = posts from 2015, after the Balmain collaboration announcement. The syntax for this procedure can be found in Appendix E.2. This variable will be used to test differences in social media reputation over time. Even though Date could have been recoded in two groups, those before and those after the announcement per year all together, it was thought best that recoding it into an ordinal variable with six distinct groups would yield more interesting results and allow for a more thorough, through time, examination between the groups compared to a binary variable, which would group all years together. Additionally, this follows the logic of yearly results, more popular in organizational reports.
Results
Before hypotheses-testing, basic descriptive statistics, were run on the sample, and the measures of centrality and dispersion for all variables are presented below (Table 1). Table 1
Measures of centrality and dispertion for all variables based on measurement level (n = 803)
Mo Mdn M SD Co-Branding 0 Relevance 1 Media 1 Co_Brand 3 Year 3 CDateA 4 Comments 556.66 684.14 Visibility 79138.74 41400.25
Favorable 43.90 14.95
Unfavorable 3.48 5.24
Neutral 36.97 16.13
Irrelevant 53.84 16.95
Favorability 0.28 0.18
Overall, in the sample (n = 803), 90.8 % of the posts were images while 9.2% were videos and 82.4% were directly relevant to the organization. Among those relevant posts, 10.5% were about one of the collaborations in question, where 23.2% regarded the collaboration with Isabel Marant, 29% with Alexander Wang and 47.8% with Balmain. More specifically, for year 2013 (n = 91) 18.7% of the total posts were about the designer collaboration, while for year 2014 (n = 250) 8% and for year 2015 (n = 325) 10.2%. As far as the variables regarding the amount of comments per valence category for each post are concerned, irrelevant comments come first, followed by favorable, neutral and finally unfavorable.
In addition, crosstabs investigating the relationship between variable Year and variables Comments, Visibility, Favorability, between variables Comments and Favorability and between variables Visibility and Favorability were also executed. Somer’s d was used as the measure of association investigating the relation between variable Year, which is ordinal, and variables Comments, Visibility and Favorability, which are ratio. Somer’s d was chosen and the relationship between these variables is assumed to be asymmetric, in terms that year influences the number of comments, likes (visibility) and the overall post favorability, and not the other way around. The results show that there is a significant, moderate positive association between Year and Comments: As year increases, the number of comments on a post increases as well, r = .43, p < .001. In addition, there is a significant, very strong positive association between Year and Visibility: As year increases,
social media visibility increases as well, r = .91, p < .001. However, there is a significant, weak negative association between Year and Favorability: As year increases, social media favorability decreases, r = -.30, p < .001. In order to investigate the relationship between the number of comments on a post (Comments) and the overall favorability of the content of those comments (Favorability), Pearson’s R was calculated since both are ratio variables. The results yielded a significant, moderate negative association between variables Comments and Favorability: As the number of comments on a post increases, the overall favorability of their content decreases, r = -.33, p < .001. Finally, Pearson’s R was calculated in order to measure the covariance between ratio variables Visibility and Favorability. These results show that there is a significant, moderate negative association between visibility and favorability: As one increases, the other decreases, r = -.43, p < .001. Differences in social media visibility between cobranding and non-cobranding related posts
The first hypothesis argued that posts regarding the yearly designer collaborations will score higher on Visibility than the rest of the organizational posts. In order to test H1A, an independent sample T-test is conducted with Visibility as the test variable and Co_branding as the grouping variable, thus comparing the two groups, co-branding related posts (coded as 1, n = 70) and non-cobranding related posts (coded as 0, n = 590) in terms of Instagram likes. The two group means come from separate groups thus an independent sample T-test is used. The t-test yielded insignificant results, t (658) = 1.35, p = .178, 95% CI [-3218.24, 17343.66]. Levene’s test was significant (p = .083) therefore equal variances for the two groups are assumed. In terms of Instagram likes, posts regarding the designer collaborations (M = 72911.94, SD = 44849.79) do not differ significantly than the rest of the organizational posts (M = 79974.65, SD = 40997.14). These results (Appendix F, Table F.1) show that there is no significant difference between the two groups of organizational Instagram posts in terms of visibility, therefore hypothesis 1A is rejected.
Differences in social media favorability between cobranding and non-cobranding related posts
Hypothesis 1B argued that posts regarding the yearly designer collaborations will score higher on social media favorability than the rest of the organizational posts. Thus, in order to test H1B an independent sample T-test is again conducted, with Favorability as the test variable and Co_branding as the grouping variable, thus comparing the two groups, co-branding related posts (coded as 1, n = 70) and non-coco-branding related posts (coded as 0, n = 590) in terms of social media favorability. The two group means come from separate groups thus an independent sample T-test is used. The t-test yielded insignificant results, t (658) = 1.23, p = .220, 95% CI [-0.02, 0.07]. Levene’s test was significant (p = .854) therefore equal variances for the two groups are assumed. In terms of Favorability, posts regarding the designer collaborations (M = 0.26, SD = 0.17) do not differ significantly than the rest of the organizational posts (M = 0.29, SD = 0.18). These results (Appendix F, Table F.2) show that there is no significant difference between the two groups of organizational Instagram posts in terms of social media favorability. Therefore hypothesis 1B is rejected. Differences in social media visibility over time
In hypothesis H2A, it is assumed that there is a difference between posts published before the announcement of the designer collaboration and posts published after the announcement for each year, in terms of visibility. As mentioned above, instead of grouping the data of all three years into two groups, forming a binary variable, an ordinal variable, namely CDateA was created. Considering more than two groups will be compared, an ANOVA will be run with Visibility as the dependent variable and CDateA as the grouping variable. On average, posts published before the announcement in 2013, belonging in group 0 (M = 15593.37, SD = 9934.15) score the lowest on Visibility, followed by posts published after the announcement in year 2013, belonging in group 1 (M = 22142.41, SD = 7926.78), followed again by posts published before the announcement in year 2014,
belonging in group 2 (M = 39785.62, SD = 12233.10), those published after the announcement in year 2014, belonging in group 3 (M = 67755.91, SD = 20999.24), those before the announcement in year 2015, belonging in group 4 (M = 102748.91, SD = 26695.45) and finally those after the announcement in year 2015, belonging in group 5 (M = 117200.39, SD = 26104.35), which score the highest (Table 2). A one-way ANOVA is carried out to assess the group differences in terms of social media visibility. The test yielded significant results, F (5, 657) = 343.41, p < .001, η2 = .72 (Table 3). A post-hoc Bonferroni test indicated that all groups had significant differences, except group 1 with group 0 (Mdifference = 6549.04, p = 1.000). The complete Bonferroni results are presented in Appendix G1, whereas the relevant to the hypothesis differences are presented in Table 4. It should be noted that the assumption of equal variances in the population has been violated, Levene's F (5, 657) = 19.08, p < .001. H2A argues that group 3 > group 2 (Mdifference = 27970.29, p < .001) and group 5 > group 4 (Mdifference = 14451.48, p < .001), which is supported by the results of the analysis, however, it also argues group 1 > group 0, which is not supported since no significant differences were found between the two groups (Mdifference = 6549.03, p = 1.000). Thus, H2A is partially supported.
Table 2
Visibility (in likes) by CDateA descriptives
n M SD
Posts before the collaboration announcement with Isabel Marant in 2013 (Group 0)
27 15593.37 9934.15
Posts after the collaboration announcement with Isabel Marant in 2013 (Group 1)
59 22142.41 7926.78
Posts before the collaboration announcement with Alexander Wang in 2014 (Group 2)
Posts after the collaboration announcement with Alexander Wang in 2014 (Group 3)
160 67755.91 20999.24
Posts before the collaboration announcement with Balmain in 2015 (Group 4)
128 102748.91 26695.45
Posts after the collaboration announcement with Balmain in 2015 (Group 5)
197 117200.39 26104.35
Note. N = 663
Table 3
ANOVA with Visibility as the dependent variable and CDateA as the grouping variable.
Sum of Squares df Mean Square F p η2
Visibility 820645303959.22 5 164129060791.84 343.41 .000 .72 Error 314010035313.59 657 477945259.23
Total 1134655339272.81 662
Table 4
Differences in Visibility between the posts published before and after the announcement of the designer collaboration per year
Mean Difference p Year 2013: after and before the collaboration announcement
with Isabel Marant (Group 1 and Group 0)
6549.03 1.000
Year 2014: after and before the collaboration announcement with Alexander Wang (Group 3 and Group 2)
27970.29 .000
Year 2015: after and before the collaboration announcement with Alexander Wang (Group 5 and Group 4)
Differences in social media favorability over time
Hypothesis H2B assumes that there is a difference in terms of favorability between posts published before the announcement of the designer collaboration and posts published after the announcement for each year. On average, posts published before the announcement in 2013, belonging in group 0 (M = 0.54, SD = 0.22) score the highest on Favorability, followed by posts published after the announcement in year 2013, belonging in group 1 (M = 0.46, SD = 0.18), followed again by posts published before the announcement in year 2014, belonging in group 2 (M = 0.41, SD = 0.21), those published after the announcement in year 2014, belonging in group 3 (M = 0.21, SD = 0.13), those before the announcement in year 2015, belonging in group 4 (M = 0.22, SD = 0.12) and finally those after the announcement in year 2015, belonging in group 5 (M = 0.23, SD = 0.11), which score the lowest (Table 5). A one-way ANOVA is carried out to assess the group differences in terms of social media favorability. The test yielded significant results, F (5, 657) = 64.75, p < .001, η2 = .33 (Table 6). A post-hoc Bonferroni test indicated that all groups differed significantly except group 1 with group 0 (Mdifference = -0.08, p = .299) and group 2 (Mdifference = 0.04, p = 1.000), group 3 with group 4 (Mdifference = -0.01, p = 1.000) and group 5 (Mdifference = -0.03, p = 1.000) and finally group 4 with group 5 (Mdifference = -0.01, p = 1.000). The complete Bonferroni results are presented in Appendix G2, whereas the relevant to the hypothesis differences are presented in Table 7. It should be noted that the assumption of equal variances in the population has been violated, Levene's F (5, 657) = 20.82, p < .001. H2B argues that group 1 > group 0 (Mdifference = -0.08, p = .299) and group 5 > group 4 (Mdifference = 0.01, p = 1.000), however the analysis did not significant differences between those groups. H2B also argues that group 3 > group 2 and even the two groups differ significantly (Mdifference = -0.21, p < .001), the analysis shows that favorability is higher for group 2 than group 3. Hypothesis 2B is thus rejected.
Table 5
Favorability by CDateA descriptives
n M SD
Posts before the collaboration announcement with Isabel Marant in 2013 (Group 0)
27 0.54 0.22
Posts after the collaboration announcement with Isabel Marant in 2013 (Group 1)
59 0.46 0.18
Posts before the collaboration announcement with Alexander Wang in 2014 (Group 2)
92 0.41 0.21
Posts after the collaboration announcement with Alexander Wang in 2014 (Group 3)
160 0.21 0.13
Posts before the collaboration announcement with Balmain in 2015 (Group 4)
128 0.22 0.12
Posts after the collaboration announcement with Balmain in 2015 (Group 5)
197 0.23 0.11
Note. N = 663
Table 6
ANOVA with Favorability as the dependent variable and CDateA as the grouping variable.
Sum of Squares df Mean Square F p η2
Favorability 6.96 5 1.40 64.75 .000 .33
Error 14.13 657 0.02
Table 7
Differences in Favorability between the posts published before and after the announcement of the designer collaboration per year
Mean Difference p Year 2013: after and before the collaboration announcement
with Isabel Marant (Group 1 and Group 0)
-0.08 .299
Year 2014: after and before the collaboration announcement with Alexander Wang (Group 3 and Group 2)
-0.21 .000
Year 2015: after and before the collaboration announcement with Alexander Wang (Group 5 and Group 4)
0.01 1.000
Differences between the three collaborations in terms of social media visibility
Hypothesis H3A argues that there is a difference in terms of visibility between posts regarding the three designer collaborations. For this analysis, the data (n = 64) are not normally distributed, the central limit theorem does not apply and it is not possible to treat the outliers, thus the analysis was conducted as is. On average, posts regarding the Isabel Marant collaboration (2013) (M = 22373.13, SD = 7486.46) score the lowest on Visibility, followed by posts Alexander Wang collaboration (2014) (M = 19453.87, SD = 4350.02) and finally those regarding the Balmain collaboration (M = 112865.03, SD = 27436.32), which score the highest (Table 8). A one-way ANOVA is carried out to assess the group differences in terms of social media visibility. The test yielded significant results, F (2, 66) = 106.88, p < .001, η2 = .76 (Table 9). A post-hoc Bonferroni test indicated that all groups differed significantly. More specifically, posts regarding the Balmain collaboration differ significantly from posts regarding the Isabel Marant collaboration (Mdifference = 90491.90, p < .001) and posts regarding the Alexander Wang collaboration (Mdifference = 62890.98, p < .001). In addition, posts regarding the Alexander Wang collaboration differ significantly from
be noted that the assumption of equal variances in the population has been violated, Levene's F (2, 66) = 7.39, p = .001. These results (Table 10) support H3A, which assumed that there will be differences between the three groups in terms of social media visibility, with the most recent one scoring the highest and the least recent scoring the lowest.
Table 8
Visibility (in likes) by Co_Brand descriptives
n M SD Isabel Marant 16 22373.13 7486.46 Alexander Wang 20 49974.05 19453.87 Balmain 33 112865.03 27436.32 Note. N = 69 Table 9
ANOVA with Visibility as the dependent variable and Co_Brand as the grouping variable.
Sum of Squares df Mean Square F p η2
Visibility 104028112266.16 2 52014056133.08 106.88 .000 .76
Error 32119368313.67 66 486657095.66
Total 136147480579.83 68
Table 10
Differences in Visibility between posts regarding the three designer collaborations
Mean Difference p
Balmain with Isabel Marant 90491.90 .000
Balmain with Alexander Wang 62890.98 .000
