Brand equity surveys or social media-based
brand equity: Which best predicts future
firm performance?
Predicting future firm performance with social networks
2
Brand equity surveys or social media:
Which best predicts future firm
performance?
Predicting future firm performance with social networks
Master’s thesis Marketing EBM867B20 Msc Marketing Intelligence
University of Groningen Faculty of Economics and Business
July 14, 2020 Ramon Vallinga Aweg 26-1 9718 CW Groningen +31 6 3159 5267 r.a.vallinga@student.rug.nl S3856739
First supervisor: Dr. E. de Haan Second supervisor: Dr. A.E. Vomberg
3
MANAGEMENT SUMMARY
For a firm, it is vital to be able to assess and compare the return on investments to enable the best possible resource allocation. Failure of accurately predicting firm performance is suggested to be due to the negligence of determining the value added for intangible assets, however, they are critical elements of firm value. In the last quarter-century, S&P companies gained a significant degree of their value due to intangible assets, while tangible assets are responsible for only a small part of this. Therefore, to be able to predict firm performance accurately, intangible assets such as brand equity should be the main metric. For these reasons, this study is focused on measuring customer-based brand equity (CBBE). Current CBBE measures utilize surveys and focus groups, which are very costly and time-consuming to conduct. Besides that, it is only measured once in a while and always looks back in time. Social media is suggested as a way to support the prediction of future firm performance and therefore enabling the best possible resource allocation. It brings the advantage of being able to monitor customers’ conversations, brand attitudes, customer engagement, and
disengagement at real-time basis, while being much more cost-effective.
The aim of this research is to evaluate the extent to which it is possible to predict future firm performance utilizing social media-based brand equity (SMBBE). This prediction on future firm performance is then compared with the future prediction of traditionally measured CBBE to decided which source best predicts future firm performance.
4 dataset contains proxies for the SMBBE dimensions in order to be comparable. Next, five performance metrics have been used to represent the firm performance, which are both forward- and backward-looking financial performance measures. Finally, the type of industry and brand likeability are measured to be able to test whether they moderate the relationship between brand equity and future firm performance.
This study has shown that SMBBE is a better predictor of future firm performance than the traditional CBBE measure. Using single SMBBE dimensions to predict future firm performance gives higher predicting power than using solely BAV dimensions. Also, combining SMBBE dimensions yields better results than combining BAV dimensions, concerning the prediction of future firm performance. Finally, four out of nine SMBBE dimensions appear to be significant predictors of a future firm performance metric. Therefore, it has shown to be a good substitute for classical survey-based brand equity in predicting future firm performance, while doing so in a quicker and more cost-effective manner.
5
PREFACE
You are now reading the thesis “Brand equity surveys or social media: which best predicts firm performance?”. I was engaged in researching and writing this thesis from February to July 2020. Even though it was difficult at times, by conducting this research I have gained a lot of knowledge on the predicting power of social media on firm performance. Besides that, I also greatly improved my R and Phyton skills. This thesis has helped me in identifying my work-related goals. I am now much more confident that I want to become a data analyst.
In truth, I could not have achieved this without the help of a few people. It gives me pleasure to be able to acknowledge those who supported me during the completion of this thesis. First of all, I wish to gratefully thank my first supervisor, dr. Evert de Haan. His ongoing support, time, enthusiasm, and his calmness in times where it was needed, were invaluable and definitely helped me through it.
I also gratefully thank Anna, my family and friends who had to cope with me at my worst moments. I thank them for their patience, support and encouragement to help me see the light again and to not give up. I hereby promise that I will never do such a thing again, at least for the near future.
Finally, I gratefully thank my fellow Master students Darius and Damianos, who helped me with finding out how to do the text analyses at times where I was flooded with errors.
6
Contents
1 INTRODUCTION ... 8
2 LITERATURE REVIEW ... 12
2.1 Brand equity ... 12
2.2 Customer-based brand equity ... 14
2.3 Existing methods to evaluate brand equity ... 14
2.4 Social media and brand equity ... 16
2.5 Twitter ... 18
2.6 Firm performance and brand equity ... 19
2.7 Model and hypotheses development ... 20
2.7.1 Brand awareness ... 20
2.7.2 Brand associations ... 21
2.7.3 Brand loyalty ... 22
2.7.4 Perceived quality ... 23
2.7.5 Brand relevance ... 23
2.7.6 Tweets on firm performance ... 24
2.8 Moderating variables ... 25 2.8.1 Type of industry ... 25 2.8.2 Brand likability ... 26 3 DATA COLLECTION ... 28 3.1 Conditions on datasets ... 28 3.1.1 BAV data ... 29 3.1.2 SMBBE data ... 29
3.1.3 Firm performance data ... 30
3.1.4 Descriptive statistics ... 31
7
4.1 From tweets to SMBBE ... 34
4.2 Sentiment analysis ... 34
4.3 Statistical models ... 35
4.4 Analysis per hypothesis ... 36
4.4.1 Brand awareness (H1a) ... 36
4.4.2 Brand associations (H1b) ... 37
4.4.3 Brand loyalty (H1c) ... 37
4.4.4 Perceived quality (H1d) ... 38
4.4.5 Brand relevance (H1e) ... 39
4.4.6 Tweets on firm performance (H2) ... 39
4.4.7 Type of industry (H3) ... 41
4.4.8 Brand likeability (H4) ... 41
5 RESULTS ... 42
5.1 Interpretation of predictions ... 42
5.2 Assumptions ... 46
6 CONCLUSION, LIMITATIONS AND FUTURE RESEARCH ... 48
6.1 Practical implications ... 49
6.2 Limitations and future research ... 50
REFERENCES ... 52
8
1 INTRODUCTION
“A brand is the set of expectations, memories, stories and relationships that, taken together, account for a consumer’s decision to choose one product or service over another.”
- Seth Godin
Research shows that analysts from all over the world are having trouble predicting the performance of firms effectively (e.g., Ferdaous and Rahman, 2019; Gupta, Lehmann & Stuart, 2004; Malkiel, 1970; 1991). Monitoring performance and adequately managing this is a primary requirement for the firm’s capacity to create and maintain long-term value for shareholders and stakeholders in today’s highly competitive market. It is of critical essence for every firm in every industry to be able to assess and compare the return on investments in order to enable the best possible resource allocation. (Schiuma, Lerro, Costa, and Evangelista, 2008).
While the economic value added to the firm for tangible assets can be calculated quite straightforwardly, it is suggested that failure of accurately predicting firm performance is due to the negligence of determining the value added for intangible assets (Lev and Zarowin, 1999; Saaty, Vargas, and Dellmann, 2003). The intangible and non-financial assets are not taken into account in the models that analysts use but, instead, the focus is on data on tangible assets (Gupta, Lehmann, and Stuart, 2004; Hogan et al., 2002). However, intangible assets are critical elements of firm value (Aaker and Jacobson, 2001; Amir and Lev, 1996; Keller 2016; Srivastava, Shervani, and Fahey, 1998). According to Gupta and Lehmann (2003, p. 10), “This interest in intangibles arises from the recognition that market value of the largest 500 corporations in the United States is almost six times the book value.” Especially in the last quarter-century, the value of companies listed on the S&P 500 has undergone a significant deviation from its book value, as shown in Figure 1 and Figure 2 (Ocean T.O.M.O, 2009).
customer satisfaction (Haskel and Westlake, 2018).
Figure 1 Value of tangible assets vs. intangible assets for S&P companies, 1975-2018 (AON, 2019)
Figure 2 Ratio of the value of tangible assets vs. intangible assets for S&P companies for the years 1975-2018
In this study, the focus will be on measuring customer-based brand equity (CBBE). The definition of Keller (2003), perceived as the most widely referenced definition of CBBE, is that brand equity is “the differential effect of brand knowledge on consumer response to the marketing of the brand.” Other researchers define brand equity as “the consumer’s response to a brand’s actions relative to competing brand’s actions” (Capon, 2013). CBBE has been discussed a lot by researchers, who have focused on its contribution to firms in terms of performance-related measures (Wang and Sengupta, 2016; Felício, Duarte, Caldeirinha & Rodrigues, 2014). The effect of CBBE on firm performance is partially analyzed by
researchers, as it appears that quality perception and brand awareness have a positive effect on return on stocks (Aaker and Jacobson, 1994). Mizik and Jacobson (2008) found that brand
17,1 32,1 68,0 80,0 84,0 83,1 68,0 32,0 20,0 16,0 1975 1985 1995 2005 2018 R at io o f va lu e o f a ss et s (i n % )
Time (in years)
Value of tangible assets vs. intangible
assets for S&P companies, 1975-2018
10
relevance and brand awareness provide incremental information to accounting measures in explaining stock returns.
The current CBBE measures make use of surveys and focus groups, which bring the disadvantage of being very costly and time-consuming (Denscombe, 2014). Thus, it is measured only once in a while and always from a historical perspective. However, online (social) media enables marketers to communicate with customers through new channels, therefore creating a brand space that customers can easily access and interact with (Srinivasan, Rutz and Pauwels, 2016). It also gives firms the advantage of being able to monitor customers’ conversations, brand attitudes, customer engagement, and disengagement more quickly and cost-effectively than by using traditional surveys (De Matos and Rossi, 2008; Pauwels & van Ewijk, 2014). What differentiates these new activity-based metrics from the traditional CBBE metrics is that they are behavior-based: they express what customers actually do. Also, they offer potentially lower tracking costs (e.g., surveys are costly) and the possibility for adaption of the firms’ strategy or campaign in an earlier stage, since the data is more fine-grained and can be accessed at real-time (e.g., surveys are, at best, executed
monthly) (Srinivasan, Rutz and Pauwels, 2016). For these reasons, the online social media platform Twitter, being one of the most popular microblogging services, is suggested as a way to support the prediction of firm performance and therefore enabling the best possible
resource allocation (Jansen et al., 2009).
This research aims to evaluate the extent to which it is possible to predict future firm performance utilizing social media-based brand equity (SMBBE) and to reduce the gap between market value and book value. SMBBE does not make use of a questionnaire or survey but uses readily available information from millions of people who have posted
something about the brand. Therefore, it is a cheap and accessible way to measure collectively build brand equity. The prediction on future firm performance is then compared with
traditionally measured CBBE, after which it can be decided which source best predicts future firm performance. To complete this objective, the following main research question is
formulated:
11
Thus far, the author has no knowledge of previous research on using SMBBE as a measurement of brand equity and linking this to future firm performance or comparing it in terms of traditionally measured CBBE.
12
2 LITERATURE REVIEW
In this section, the definitions of brand equity, CBBE, social media, and firm performance are provided. Subsequently, existing brand equity measures are reviewed, followed by the characteristics of the chosen social media platform, which will be used to measure SMBBE, namely Twitter. Next, the current understanding of the relationship between firm performance and brand equity is evaluated. After that, this study’s hypotheses about the relationships between tweets, brand equity, and firm performance with moderating variables, accompanied by arguments, are presented. Finally, these relationships are
visualized in a conceptual model.
2.1 Brand equity
Brand equity is defined in multiple ways, and there is no standard way of measuring it. Brand equity can be defined as “outcomes that accrue to a product with its brand name compared with those that would accrue if the same product did not have the brand name” (Ailawadi, Lehmann, and Neslin, 2003, p.1). That is, the gains that a product has because of the power that comes with its brand name. Brand equity can both enhance and reduce this power and thus consist of brand assets and liabilities. The received balance sheet valuation of a brand when a firm is acquired is an example of evidence that a brand has a value that can be measured (Bahadir, Bharadwaj & Srivastava, 2008). The value associated with the acquisition of one company by another is referred to as goodwill. Examples of this intangible asset are the value of a company’s brand name, customer base, and customer and employee relations (Investopedia, 2020). This evidence is further supported by an example of the substantive sale price of a brand even after bankruptcy and liquidation of a firm’s remaining assets (Zipkin, 2009). The importance of measuring the impact of brand equity from not only the standpoint of managers and marketers but also for firms who want to value their worth and potential, many attempts have been made to measure brand equity accurately. According to Keller and Lehmann (2003), these attempts can be categorized into three distinct approaches:
• customer mind-set or attempts to measure brand equity from the consumer’s point of view, including measures such as awareness, attitudes, or attempts and associations (e.g., Aaker, 1996; Hupp and Powaga, 2004; Keller, 2008). According to the
13
• product-market outcomes, such as a measure of price premium or the difference in the price commanded by a brand-name product versus a generic product (e.g., Aliwadi, Lehmann, Neslin, 2003; Park and Srinivisan, 1994). As stated in the framework of Gupta and Zeithaml (2006), this approach describes the relationships between what customers think and what customers do;
• financial-market outcomes, such as an association between brand equity volatilities and stock returns (e.g., Barth, Clement, Foster & Kasznik, 1998; Wang, Zhang & Ouyang, 2009; Kaparelitois and Panopoulos, 2010). This financial brand equity is likewise named brand value as it connects a monetary value to the brand. According to the framework of Gupta and Zeithaml (2006), this approach captures the relationship between what customers think and what firms get
Figure 3 Framework for customer metrics and their impact on firms' financial performance (Gupta & Zeithaml, 2006)
These approaches all have their strengths and weaknesses (Ailawadi, Lehmann, and Neslin, 2003). While financial market measures capture both present and prospective
potential, they often rely on subjective judgments or inconsistent measures to estimate value (Simon and Sullivan, 1993). Product market measures appear to be relatively more closely related to marketing activities. However, they are not able to capture prospective potential. Although financial market and product market measures provide insights for strengthening brand equity, they give not much information about brand performance in terms of
14
bypass unobserved metrics.” When these quantitative modelers assess the short-term and long-term sales and profit effects of the marketing mix, they regard the customer’s mind and heart as a “black box” (Hanssens, Parsons & Schultz, 2003). Mind-set metrics are typically used by branding and advertising experts and researchers in consumer behavior, who aim to assess the effect of marketing efforts on the customer mind-set. Thereby they often do not examine the ultimate influence on sales and also ignore the influence of competitive actions. For these reasons, Keller (2008) and this study focus on measuring brand equity utilizing customer mind-set metrics and, after that, assessing the effects this brand equity on firm performance, which Keller refers to as “customer-based brand equity”.
2.2 Customer-based brand equity
Customer-based brand equity (CBBE) is used as one of the most prominent descriptions of brand equity. Researchers adopted the view that if a firm wants to achieve long-term effectiveness of its marketing programs on the firms’ revenue and customers, building strong brand equity is a very feasible strategy to adopt (Mackay, Romaniuk & Sharp, 1998; Simon and Sullivan, 1993). Keller (2008) had the vision that the objective of CBBE was to harness the power of the brand through consumers’ responses to marketing programs. He has developed a theory based on two crucial elements: (1) awareness and familiarity and (2) strong, favorable brand associations. Thus, any customer mind-set measure of brand equity must include both elements, awareness/familiarity, and brand associations. Another well-known and well-cited definition is Aaker’s (1991). He adopted a multi-component approach in knowing, differentiating and distinguishing brands and products that are build up of mental assets and liabilities. This approach is a model consisting out of five constructs that influence CBBE, which are: “brand loyalty”, “brand awareness”, “perceived quality”, “brand
associations”, and “other brand assets”. Aaker’s CBBE approach is used by multiple scholars in their studies (e.g., Prasad and Dev, 2000; Yoo and Donthu, 2001; Pappu, Quester & Cooksey, 2005; Buil, de Chernatony & Martinez, 2008).
2.3 Existing methods to evaluate brand equity
15
models are better than financial models, yet the latter seems to be somewhat of a more static approach and are more limited as brand management tools (Zimmerman, Klein-Bölting, Sander & Murad-Aga, 2001). The focus on financial data is more useful for accounting purposes or when a brand is put up for sale since the output produced by these models is the brand value from an accounting perspective (Murphy, 1989). They do not take into account the role of the consumer in the assessment of brand equity and, therefore, aspects that are perceived as interesting from the perspective of brand managers such as brand awareness and loyalty are left out. Subsequently, these methods are unable to predict a brand’s performance in the long run since it is not possible to assess the potential of a brand if relevant information is not taken into account.
To overcome the limits of financial-based measures, several mind-set measures of brand equity have been developed which are based on behavioral science on are focused on
customers. The commercial measures such as Millward Brown’s BrandZ, Young &
Rubicam’s BrandAsset Valuator (BAV) (2000), or Research International’s Equity Engine assess four to five major facets of brand perceptions, of which the foundation is laid by Aaker (1991). Aaker (1991) focused on measuring brand equity by the five dimensions brand
awareness, brand associations, brand loyalty, perceived quality, and other proprietary brand assets. Of these commercial measures, the dynamic marketing-based evaluation technique BAV is best-known as a measure for brand equity and is “the world’s largest database of consumer-derived information on brands” (Keller, 2008 p. 393). Lehmann, Keller, and Farley (2008) show significant correlations between the BAV pillars and dimensions of other
commercial equity measures as well as measures such as attitude and satisfaction (Stahl, Heitmann, Lehmann & Neslin, 2012). Young and Rubicam has assessed brand equity for more than twenty years and has collected data of about 50.000 brands in 51 nations. Their BAV measure builds upon four ‘pillars’, that assesses the awareness/familiarity and brand association components as developed in Keller’s (2008) theory:
• Knowledge: The extent to which customers are familiar with the brand;
• Relevance: The extent to which customers perceive the brand to be relevant to their needs;
16
• Differentiation: The extent to which the brand is perceived as unique, distinct, or different.
The knowledge pillar captures the awareness/familiarity component, while the other three – relevance, esteem, and differentiation – tap brand associations. Nevertheless, also the marketing-oriented and behavioral models incorporate a few shortcomings other than the apparent advantages (they clarify what goes on in the “hearts and minds” of consumers). It appears that there is a considerable degree of subjectivity involved in the choice of factors used to explain brand equity, and they are rarely suitable for valuing different types of brands (corporate or product brands) (Zimmerman et al., 2001). Also, Young and Rubicam’s
consumer-derived information on brands is attained by surveys. Collecting data via surveys is a method that bears the disadvantage of being very costly and time-consuming and therefore, can only be conducted once in a while. Furthermore, surveys have a significant degree of bias (Denscombe, 2014). Thus, data readily available, which can easily be accessed and being inexpensive at the same time, might be an innovative, sustainable way to tackle these problems.
In the current study and aligned with Narteh (2018), brand equity will be conceptualized using a combination of Aaker’s (1991) model and Stahl et al., ’s (2012) modification of Young and Rubicam’s brand equity assessment measure. Brand equity would, therefore, be measured by brand awareness, brand associations, brand loyalty, perceived quality (Aaker, 1991) as well as relevance (Stahl et al., 2012; Young and Rubicam, 2000). In total, five variables will measure brand equity; these variables are discussed in the following sections.
2.4 Social media and brand equity
Social media can be defined as “online applications and platforms which aim to facilitate interactions, collaborations and sharing of content” (Richter & Koch, 2007). It is also characterized as “any website which allows users to share their content, opinion, views and encourages interaction and community building” (Neti, 2011). The term social media is constituted out of two words, namely ‘social’ and ‘media’. ‘Social’ indicates the
17
tool (Neti, 2011). Internet-based technology is used for the sharing of knowledge and information to a large number of users, at high velocity. It allows the creation and exchange of user-generated content. The social networking platforms are an important source of customer specialization by giving people a space to communicate through the internet (Vinerean, Cetina, Dumitrescu & Tichindelean, 2013). While the main focus of social media is creating and maintain interpersonal relationships, it is found that social media can provide social capital to firms, which helps in collecting marketing intelligence and identifying opportunities (Gillin and Schwartzman, 2011). Furthermore, it helps in monitoring the target market and customer behaviors on social media platforms (Gillin and Schwartzman, 2011). Finally, consumers are increasingly looking to a brand’s social media presence to form judgments about the brand (Baird and Parasnis, 2011; Naylor et al., 2012).
Marketing literature commonly classifies social media into owned social media (OSM) and earned social media (Srinivasan, Rutz, and Pauwels, 2016; Stephan and Galak, 2012). OSM is referred to as a brands’ communication, made and shared through its own online social network assets, for example, the brands’ Facebook page and YouTube channel. Conversely, ESM alludes to the brand-related content that entities other than the brand – in general, the customers – create, consume, and disseminate through online social networks. In this study, the focus will be on measuring ESM since the content generated, read, and spread by the customer is behavior-based. Therefore, it shows how the customer actually perceives the brand rather than how the firm desires to be viewed, which is the case with OSM.
18
Facebook like for the studied brand, in terms of sales. This shows that one should be careful in excessively relying on ESM versus OSM.
2.5 Twitter
Launched on July 13, 2006, Twitter is an American microblogging and free social networking service that allows for reading, posting, liking and retweeting of short messages, called tweets, to a network of associates, a.k.a. followers. These messages are limited to 280 characters (it used to be up to 140 characters until November 2017), and users are also able to post photos or short videos. Tweets are posted to a publicly available profile or can be sent as direct messages to other Twitter users (Statista, 2019). Even though analyses on Twitter data have not been executed much in the marketing literature (except for Toubia and Stephen (2013) and Stephen et al., (2010)), it is decided that Twitter is a good social networking site for our analysis. First of all, it is relatively popular. As of the first quarter of 2010, Twitter averaged 30 million monthly active users. That number has since then increased to 330 million monthly active users by the first quarter of 2019 (Statista, 2019). This huge increase makes exploiting value out of tweets nowadays even more useful. By mid-2013, 77% of Fortune 500 firms had active Twitter accounts, which is more than the 66% that had Facebook pages (Barnes et al., 2013) and 62% that have YouTube channels (Heggestuen and Danova, 2013). Furthermore, the platform is relevant for this study. Research shows that 42% of Facebook users have ever mentioned a brand in their status update (Mazin, 2011), while about 19% of all tweets by Twitter users are brand-related (Jansen et al., 2009).
Twitter is used for creating and maintaining brand image and personality development, as frequent and conversation-like messages can be conveyed at low cost to an enormous brand community (Etter and Plotkowiak, 2011; Kim and Ko, 2012; Kwo and Sung 2011). Accounts can be held at the firm or brand level, permitting communities to develop at scale fitting to a firms’ brand strategy – a critical distinction when researching brand image perceptions, as brands can be dominated (to varying degrees) by their parent corporate brands (Berens et al., 2005). Third, its social connections are publicly visible, aside from a little minority of protected accounts, which were estimated at about 8% in 2009 (Cha et al., 2010). At last, Twitter is organized. Since Twitter accounts are usually organized by users into topic-based lists, accounts users consider applicable to a perceptual attribute that can be classified
19 to measure SMBBE for the reasons mentioned above, it could also be extended to other platforms, and future research is encouraged to do so.
2.6 Firm performance and brand equity
According to the efficient market hypothesis, a firm’s stock price or firm valuation always, consistently expresses what is known by investors and prospective investors (Fama, 1991). Said differently, the stock price of a firm indicates investors’ views on the current and prospective gains of all its assets, both intangible and tangible. Tangible assets incorporate the firms’ property, plant, and equipment, current assets such as inventory and investments. These assets are predominantly measured by the assets’ replacement cost (Simon and Sullivan, 1993). Tangible assets, on the other hand, include any other asset that may empower a firm to gain excess returns, past that earned from its tangible assets. It incorporates, for example, patents and trademarks, investments in R&D, altruism, and, as introduced here, brand equity (Simon and Sullivan, 1993). As indicated by Simon and Sullivan (1993, p. 31), “the financial markets view brand equity as the capitalized value of the profit that results from associating the brand’s name with particular products or services”. Therefore, anything that may affect investors’ perceptions of brand equity, let it be negative or positive, should influence the firm’s stock price because of the view of the impact on future gains. For instance, if a firm attempts a very important promotional campaign and investors think it will be effective, they will probably drive the price of the stock, equivalent to the variation between the cost of the campaign and possible increment of the firm’s gains.
It is worth mentioning that the accounting effect of such a campaign will take place in the year in which the campaign is launched. However, the effect on brand equity could possibly be felt for a long time after, therefore potentially influencing the earnings for numerous years to come. Similarly, an unfavorable event such as Toyota’s brake failures issues in 2010 can have a significant negative effect on a firm’s stock price because of perceptions of a negative long-term effect on a brand, well beyond the immediate effect on sales of that year. Other scholars suggest that firm performance can be measured in terms of profits and returns on investments (ROI) (Aaker and Jacobson, 2001).
20
between online and offline consumer loyalty (Danaher et al., 2003) and brand image (El Gazzar and Mourad, 2012). Brand equity elements measured whose changes appear to affect firm valuations incorporate brand attitude (Aaker and Jacobson, 2001), customer satisfaction (Fornell, Mithas, Morgeson & Krishnan, 2006), new product introductions (Pauwels, Silva-Risso, Srinivasan & Hanssens, 2004) and perceived quality (Aaker and Jacobson, 1994). Brand orientation had additionally been demonstrated to be related to firm profitability (Gromark and Melina, 2011), and advertising expenditures positively affect the intangible value of a firm (Sahay and Pillai, 2009). Possibly most extensively, Mizik and Jacobson (2008) show a quantifiable effect of changes in perceived brand relevance and energy, and also a time-lagged effect of brand awareness on stock returns.
2.7 Model and hypotheses development
The conceptual model is shown in Figure 4. It assumes that tweets are able to (a) measure brand equity and its associated five dimensions and (b) predict future firm performance. This relationship is moderated by the type of industry and brand likeability. Each component of the model and its related hypothesis are discussed in the subsequent sections.
Figure 4 Conceptual model of the prediction of SMBBE on future firm performance, including the moderating variables “type of industry” and 'brand likeability'.
2.7.1 Brand awareness
21
(2013) furthermore suggests that brand awareness is important in the process of decision making of customers since it supports them in making them conscious of the availability of a given brand and preferably in a given product category. Since brand awareness is composed of brand recognition and brand recall (Keller, 2013), brands that are capable of creating a significant level of awareness amongst their consumers will increase sales, which will, in turn, increase profit and thus positively affect future firm performance. Therefore, the following hypothesis is proposed:
H1a: The volume of tweets related to a brand in a given period is positively related to
self-indicated brand awareness for that brand in the future period.
2.7.2 Brand associations
The objective of each firm should be to create favorable brand associations (Keller, 2008). These have been defined as “anything linked in memory to a brand” (Aaker, 1991, p. 109), what’s more, the more customers associate certain encounters with a brand, the more grounded the brand association will be with that specific experience or cue (Aaker and Keller, 1990). Customers form perceptions based on their experiences with the brand, its marketing activities, employees, and product performance. Brands that are able to satisfy consumers’ wishes will succeed with regards to making strong brand associations that support consumers’ positive perceptions of the brand and therefore, will be protected against competitive actions (Narteh, 2018). As can be imagined, brand associations are affected by lots of items such as social image, trustworthiness, distinctiveness, organizational associations, and even country of origin (Fayrene and Lee, 2011). Research shows that brand associations are very important in building brand equity (Del Río, Vazquez & Iglesias, 2001), as they serve as the basis for customer purchases, by helping process, organize and retrieve information within their memories (Aaker, 1991; 2009; Low and Lamb Jr, 2000). Keller (1993) affirmed this when he suggested that there are three kinds of brand associations that result in the development of brand equity: attributes, benefits, and attitudes. Attributes can be distinguished by how directly they relate to product/service performance and appearance. Benefits refer to the meaning that customers attach to the product or service, while brand attitudes are defined in terms of customers’ overall evaluation of the brand during the customer journey (Brandvision, 2009). Therefore, if a brand is able to effectively create beneficial associations with its
22 enhance the financial performance of a firm (Aaker and Jacobsen, 2001). Thus, the following hypothesis is made:
H1b: The share and sentiment of tweets related to brand associations for a brand in a given
period is positively related to self-indicated brand associations towards that brand in the future period.
2.7.3 Brand loyalty
Keller (2008) has defined brand loyalty in terms of resonance: that degree of customer-brand relationship which represents a match up between the brand and its
customers, and which produces unusual behavioral outcomes, for example, customers actively looking for intends to interact with and share their brand encounter with others. Aaker (1991), on the other hand, defines brand loyalty as a circumstance that reflects how likely a customer will be to switch to another (substitutable) product of a different brand. This person tends to be more likely to switch when that brand makes a change, either in price or in product features. Loyalty can solidly be put as a part of brand equity as, when customers respond more strongly to a brand’s actions as compared with competing brands, sales and profitable returns on investment will increase (Capon, 2013). Besides, it appears that brand loyalty affects firm performance (Keller, 2003). Brand loyalty reduces marketing costs in two ways: (i) retaining customers is overall less costly than attaining new customers, and (ii) a satisfied customer base is likely to attract new customers on its own (Aaker, 2009). Brand loyalty is not only assessed by means of customer satisfaction upon previous encounters; it is also influenced by brand trust, which is found to be the key construct of brand loyalty. Brand trust is “the feeling of security that the brand will meet the customer’s expectations” and has been found to be positively correlated with brand loyalty (Delgado-Ballester and Munuera-Alemán, 2001). Through brand loyalty, firms increment their overall revenues while moreover
leveraging the brand and setting up measures to bear competitive actions. Furthermore, research suggests that brand loyalty gives significant competitive and economic benefits to a firm. Without loyalty, firms cannot accomplish a sustainable competitive advantage
(Delgado-Ballester and Munuera-Alemán, 2005; Chaudhuri and Holbrook, 2001). Therefore, the following hypothesis is proposed:
H1c: The proportion and sentiment of tweets related to brand loyalty for a brand in one period
23
2.7.4 Perceived quality
When discussing the concept of brand equity, perceived quality cannot be left out. Aaker (1991, p. 85) defined perceived quality as “the customer’s perception of the overall quality or superiority of a product or service with respect to its intended purpose, relative to alternatives.” It is a critical component of brand equity and reflects how customers’
perceptions of a brand’s quality impact their purchase decisions. When customers consider a brand to be of outstanding quality, they will like it and participate in repeat purchases and positive word of mouth (Kotler and Armstrong, 2010; Keller, 2013). It does not automatically represent the actual quality of the product or service, but it is instead based upon the users’ subjective evaluations (Yoo and Donthu, 2001). Unlike perceive quality, objective quality does not necessarily contribute to brand value (Anselmsson, Johansson, Persson, 2007). The concept depends in general upon the product’s or service’s reliability, durability,
serviceability, style, and design. Consumers’ perceptions related to these characteristics often define quality and therefore affect their attitudes and behaviors towards the brand (Keller et al., 2011). According to Vantamay (2007), the determinants of perceived It is suggested that there is a link between product and service quality, customer satisfaction and firm profitability (Kotler, 2000), while other researchers found a direct relationship between a firms’ quality management orientation and financial performance (O’Neill et al, 2016). Furthermore, perceived quality is found to be associated with price premiums, price elasticities, and even more important, stock returns. Based on this, the third hypothesis is proposed:
H1d: The amount and sentiment of tweets related to perceived quality for a brand in a given
period is positively related to self-indicated perceived quality towards that brand in the future period.
2.7.5 Brand relevance
Despite the fact that brand relevance has traditionally not been included as a component of brand equity, Stahl et al., (2012) incorporate it as a measurement and
24 the consumer market is nowadays so dynamic that brand categories are made but become dim at a brisk pace. Brands should thusly continually develop so as to stay relevant to their
consumers and therefore be successful. The significance of brand relevance to brand equity has been observed, as the differentiation of a brand will just matter when it is considered to be relevant by the customer (Mizik and Jacobsen, 2008; Young and Rubicam, 2000).
Furthermore, Young and Rubicam (2000), who characterize the relevance of a brand as a pillar that adds to the overall strength of the brand, suggest that if the brand has low relevance to customers, it will not be highly favored by them. These suggestions have been strengthened by Keller (2013) and Park et al., (1986), who indicate that the brand creates relevance through functional, experiential, or symbolic benefits. Roll (2009) has the view that firms must invest in creating brands that can leverage on their core promises and doing as such, deliver relevant products and services that will fulfill customers. As a result of this, a high level of
commitment to the brand will be developed. Guzmán, Abimbola, Tolba & Hassan (2009) repeated the significance of relevance as a determinant of brand equity. They found that brand equity is positively linked with a firm’s performance. In view of this and to contribute to the completeness of the predicting model, decided is that this study will include brand relevance as a component of brand equity. Therefore, the following hypothesis is proposed:
H1e: The share and sentiment of tweets related to brand relevance for a brand in a given
period is positively related to self-indicated brand relevance towards that brand in the future period.
2.7.6 Tweets on firm performance
After having derived the different components of brand equity for brands by means of tweets on a given brand, the relationship between these components and future firm
25
H2: The components of brand equity, measured by tweets, are positively related to
future firm performance.
2.8 Moderating variables
Research suggests that there are some variables that moderate the relationship between brand equity and future firm performance: type of industry (Krishnan and Hartline, 2001) and brand likeability (Narteh, 2018). The effects of such are evaluated in the coming sections, after which hypotheses are developed.
2.8.1 Type of industry
Research on the effect of brand equity on firm performance for different types of industries shows that, within the service industry, brand equity has a significant effect on brand preference and purchase intentions (Chang and Liu, 2009). Successfully tested within the hospitality industry (Boo, Busser & Baloglu, 2009; Konecnik and Gartner, 2007;
Kayaman and Arasli, 2007), the effect of CBBE on firm performance has also been shown in the financial services sector. Furthermore, it has been found that customers involve in
switching between brands in the financial sector and that their arose perception of the brand can help predict loyalty intentions (Taylor et al., 2007). Additionally, research has shown that when private, foreign, and state banks in Turkey were compared, overall CBBE was greater for private banks than state banks and foreign banks (Pinar, Girard & Eser, 2012). It was pointed out that this was mostly due to poor perceived quality of foreign banks. The vast majority of research to date with respect to the financial effect of brand equity metrics has been restricted to large multinational consumer firms where one brand (or a small number of brands) accommodates most of the firm’s sales.
Some researchers have argued that strong brand equity is more essential for service-oriented firms than product-service-oriented firms. For example, Onkvisit and Shaw (1989) suggest that branding is critical since many services are regarded as commodities by customers. What’s more, the intangible nature of services impedes the evaluation of their quality for customers. Branding a service can aid customers by supporting them in assuring them of a uniform level of service quality. Branding also helps the service provider by increasing the service level above the commodity level, which differentiates the service relative to
26 acquiring a service (Bharadwaj et al., 1993). Since services possess unique characteristics, customers have trouble judging the content and quality of service, preceding, during, and after the utilization of the service (Darby and Karni, 1973; Nelson, 1970). Therefore, brand names can help to reduce the risks associated with the purchase and consumption of numerous services (Bharadwaj et al., 1993). To investigate whether brand equity really is more
important for services than for goods, a more recent study showed that this is actually not the case. They did not found consistent differences between the types of industries and even found that brand equity is more important for tangible goods than services in most cases (Krishnan and Hartline, 2001).
In light of this, this inconsistent variation between the effect of service- and product-oriented brands on customer decision making implies that the effect of brand equity on future firm performance is moderated by the type of industry. Based on this, the following
hypothesis is proposed:
H3: The type of industry (service- or product-oriented) has a moderating role in the
prediction of future firm performance by brand equity
2.8.2 Brand likability
The goal of each firm is that its customers like and disparage their products and service (Kotler and Armstrong, 2010). Brand likeability can, therefore, be viewed as an important variable that can be utilized to establish the power of the relationship between brand equity and future firm performance. To date, researchers have attempted to comprehend the role of consumer emotions in purchase decisions, resulting in studies on concepts such as brand likeability (Nguyen, Choudhury & Melewar, 2015). Brand likeability can be explained by the concepts of “attractiveness, credibility, and expertise in order to create attachment and love by delivering beneficial outcomes for consumers and brands alike” (Nguyen et al., 2013, p. 372). Therefore, it has been conceptualized as comprising the three elements of
attractiveness, credibility, and expertise (Nguyen, Melewar & Chen, 2013). Suggested is that customers are, as a matter of fact, influenced by how attractive the brand and its marketing is in combination with their attitude towards its quality and superiority (Landwehr, McGill & Herrman, 2011). Marketers have tried to induce the feeling of likeability through
27 and that this adoration serves as a foundation of their purchase decisions (Carroll and Ahuvia, 2006).
Furthermore, Batra, Ahuvia & Bagozzi (2012) suggest that customers who have an extraordinary liking for brands are important assets for the firm as they are defined by brand advocacy and evangelism as well as continuous purchases of the firms’ products and services through the span of their lifetime. They also found that these customers develop resistance to competitor advances. Taking this into account, the following hypothesis is proposed (Narteh, 2017):
H4: Brand likeability moderates the relationship between brand equity and future firm
28
3 DATA COLLECTION
This chapter starts with reviewing the conditions that have been placed on the firms to be incorporated in the datasets. Then, the three data sources used for this study are discussed separately in the following sections, according to the structure of De Haan (2020), namely (1) the BAV data of Young and Rubicam, (2) the SMBBE data scraped from Twitter and (3) the firm performance data collected from Yahoo Finance and YCharts. Also, descriptive statistics are provided of the final dataset.
3.1 Conditions on datasets
Each dataset covers a time period of two and a half years, specifically from the first quarter of 2008 until and including the second quarter of 2010, in which large U.S. brands are included. The following conditions have been placed on the firms in order to be incorporated in the datasets:
• Obviously, the firm should be covered in the BAV dataset;
• The firms in the dataset should be publicly traded in order to link the brand equity to readily available firm performance data, which will be elaborated on in section 3.1.3;
• The firm should meet the requirements of being a mono-brand firm (De Haan, 2020). A monobrand is defined as a firm “in which a single brand represents the bulk of [its] business.” (Mizik and Jacobson, p. 20-21, 2008). According to Tamrakar, Pyo, and Gruca (2018), limiting the dataset to monobrand firms is favorable, relating to the extent to which firms focus on their corporate branding. For monobrand firms, any attitude expressed about the brand online could possibly influence the entire firm’s financial performance.
• The firm should have been surveyed in all of the ten quarters of Q12008 – Q2010 in the BAV dataset, in order to increase validity;
• The firm’s account should have been mentioned in at least 200 tweets in every quarter, for all of the ten quarters in order to have enough diversity;
• Financial services sectors are excluded from this study due to their unique balance sheets and unique financial performance indicators. Thus they cannot be compared to firms from other industries (Aydin & Ulengin, 2015);
29
3.1.1 BAV data
This study uses a 2008-2010 Y&R BAV dataset as the classical CBBE measure to predict future firm performance. The BAV consulting group of Young and Rubicam conducts a quarterly survey among a representative panel of the U.S. population (17.000 people, every respondent answers on 250 brands each quarter). The survey measures a broad set of attitudes and perceptions for a large number of brands. The used data set includes information for quarter 1 of 2008 to quarter two in 2010 on 697 major U.S. brands (both corporate and
product), which are classified into 16 product categories. The data set also contains a separate file, including the quarterly information for each quarter (2008Q1-2010Q2). While a few brands were measured in all of the quarters, some brands, unfortunately, were not – e.g., Kleenex was only measured in five out of ten quarters. Not only the values of the specific questions in the Y&R BAV survey are included, but also of the four pillars of CBBE that Y&R BAV constructs from them, as discussed in section 2.3 (Lovett, Peres & Shachar, 2014).
The choice for using Y&R BAV data as the traditional measure for CBBE is fivefold: (1) the data are free and publicly available, (2) BAV is best-known as a measure for brand equity (Keller, 2008) (3) a large number of (major) U.S. firms are covered (4) the survey panel is representative of the U.S. population (5) the feasibility of predicting future firm performance by making use of BAV as a measure of CBBE has already extensively been proven by other researchers (e.g., Mizik and Jacobson, 2008; Aydin & Ulengin, 2015; Stahl et al., 2012).
After applying the seven conditions on the BAV dataset, the final sample included 20 firms that met all the criteria resulting in 200 observations. In total, the firms are represented in roughly six categories. From this point on, only data on these firms will be utilized for analyzing tweets and eventually predicting future efirm performance. In Table 1, a list with the distribution of brands across the categories is shown. In Table 2, an overview of the brands included in the dataset is displayed.
3.1.2 SMBBE data
30
the tweets in which the brand is mentioned in general (not those with @username). This decision is based on the fact that (1) almost none of the brands had initiated a Twitter account in 2008Q1, so there are simply too few tweets in which the account is mentioned to make meaningful inferences. Also, (2) it cannot be said with certainty that the firm’s account mentioned in a tweet really is about that firm since the accounts were not verified back then, and someone else could have used the username. The Phyton package Twitterscraper
developed by Taspinar (2019) was, as the name suggests, used for scraping the tweets in which the firms are mentioned in the timeframe January 1st of 2018 until and including June 30th, 2010. This timeframe is divided into ten quarters in order to be comparable with the brand equity derived from the BAV dataset. To further increase the comparability, only tweets on the firms included in the BAV dataset are collected. This resulted in a total of 314792 number of tweets, with an average of about 15740 tweets per firm.
3.1.3 Firm performance data
There are lots of ways to measure firm performance, but they can be broadly classified into two categories: accounting performance, a backward-looking performance measure, and market performance, a forward-looking performance measure. To assess brand equity’s relation with both forward- and backward-looking financial performance measures, various measures of performance from both categories will be used for the first quarter of 2008 until and including the second quarter of 2010 (Nabod, 2017).
To assess the firm’s market performance, stock performance data from Yahoo Finance is obtained, which is (1) market value (in millions of dollars), (2) trading volume (in millions of shares traded), and (3) stock return (in percentages). Stock prices are considered to
completely represent all relevant aspects of a firm’s performance and any anticipation on future performance, that is, they reflect any available information (Lubatkin and Shrieves, 1986). Stock returns characterize the fluctuation (in this case quarterly) in stock price and thus firm performance. A positive fluctuation can reflect the growth and a bright future outlook, while a negative fluctuation may mirror the opposite (Nabod, 2017).
31
firm’s invested capital has generated earnings (Nabod, 2017). ROE, characterized as the net income over total shareholder’s equity, is another traditional accounting measure that reflects the firm’s profitability. Data for these accounting performance measures are obtained from YCharts. The two measures are based on historical information and provide objectivity
(Nabod, 2017). These five firm performance indicators are established for each brand in every quarter, after which they are matched to the BAV and SMBBE data.
3.1.4 Descriptive statistics
An overview of the 200 firm-quarter observations is shown in Table 4. There is data for all predictors of future firm performance – that is, the number of tweets collected per quarter, the scores for the dimensions brand awareness, brand associations, brand loyalty, perceived quality, and brand relevance for both SMBBE and BAV data. For the variables which have observations <200, such as SMBBE relevance and SMBBE loyalty, there are actually no missing observations. Rather, the score for that firm-quarter observation was found to be 0, and therefore these observations are not treated as NA’s. Also, most of the data of the firm performance measures are available too. For some cases in which there is missing data, i.e., the ROE is not available for all firm-quarter observations, the models are estimated by using this limited data set.
32
quality and likability ratio, which was also found by Narthe (2018), rather the coefficient estimated in this study appears to be higher for perceived quality (0.879>0.441). This may imply that in an online setting, associations are more driven by perceived quality relative to an offline setting. In line with Keller’s framework (1993), this could be related to higher brand attributes.
Furthermore, for the SMBBE dimensions associations, loyalty, and perceived quality holds that there is found a positive correlation between the ratio variable and their associated net sentiment variable. This means that, for these variables, there exists a moderately strong positive coherence between the number of tweets related to such a dimension and the extent to which tweets are found to be positive/negative. For the other SMBBE dimensions, this effect is not established. SMBBE likeability ratio correlates highly on the SMBBE dimensions associations ratio, loyalty ratio, and relevance ratio. This correlation is also supported by the results of Narteh (2018).
For the BAV dimensions, it appears that BAV awareness negatively correlates with SMBBE brand awareness net sentiment, apart from the positive correlation with the ratio part, that has been mentioned before. This is not as expected since it implies that higher BAV awareness coherences with a more negative sentiment for that same dimension, but measured in an online setting. Furthermore, most of the BAV predictors correlate strongly with other BAV predictors. Similar results as with the SMBBE predictors are found here as well as awareness, associations, loyalty, relevance, and to a lesser extent, perceived quality correlate highly on each other. The signs are as expected as they are solely positive; therefore, one dimension may strengthen another, e.g., higher awareness results in higher loyalty and associations. These effects have also been established by other researchers (Nabod, 2017; Mizik and Jacobson, 2008; Kirk, C. P., Ray, I., & Wilson, B., 2013).
The bottom half of the table also shows very few unexpected results. Especially market value shows some predictors to be correlated with each other, even though the correlation is rather low. The signs are, as they are expected to be, mostly positive (Roll, 2009; Tolba and Hassan, 2009). However, the sign of the SMBBE relevance ratio is negative. Having a higher market cap usually results in higher relevance since it is linked to higher advertising
33
effective, especially when a sufficient market presence is established (Riezebos and Riezebos, 2003). Therefore, firms may be reluctant to engage in such activities, which causes the
relevance to decrease when the market value is increased. Trading volume correlates moderately on every BAV dimension except for brand loyalty. Apart from the latter, these results resonate with i.e., the results of Narteh (2018), Stahl, et al., (2012), Mizik and Jacobson (2003 and Aydin and Ulengin (2015). An explanation why brand loyalty is not found to be significantly correlated with trading volume is that research has revealed that individual investors indicated they would buy competitive offerings, suggesting that stock ownership is more likely to lead to repeat purchase behavior, but not brand loyalty
34
4 METHODOLOGY
In this chapter, the methodology of converting the tweets to SMBBE data is reviewed, after which the utilized analyses, that is, topic classification, sentiment analysis, and correlation test, are discussed. Then, the analyses per hypothesis are discussed.
4.1 From tweets to SMBBE
After having collected tweets in which the brand is mentioned for every quarter from 2008Q1 until and including 2010Q2, the tweets were converted into SMBBE data in order to be compatible with analyzing. Text analyses have been performed on the data so as to obtain values for every dimension of SMBBE for every brand in every quarter. In the following sections, the analyses will be elucidated, after which the analyses per hypothesis are
discussed. Lastly, the methodology of assessing the relationship between the predictors and future firm performance is provided.
4.2 Sentiment analysis
Sentiment analysis is the classification of emotions (positive, negative, and neutral) to identify customer sentiment towards brands in online conversations within text data using text analysis techniques (Pascual & Wolff, 2019). In this analysis, sentiment polarity scores will be given to each tweet to determine the sentiment of the tweet. Sentiment polarity scores are obtained by utilizing the R software package “sentimentR” (Rinker, 2018), and it shows the degree to which a text is positive (or negative). The algorithm of this package tags positive and negative polarized words for every tweet, using the sentiment dictionary of Matthew Jockers (2017). It then assesses the context in which the polarized words are used by taking a cluster of four words before and two words after the polarized word, which is referred to as “the context cluster”. This context cluster can exist out of five types of words (De Haan, 2020):
• Neutral words, which do not influence the meaning of the polarized word;
• Negating words, which changes the polarized word into negatives (e.g., “I do not”);
• Amplifying words, which strengthen the meaning of the polarized word (e.g., “I highly trust”);
35
• Adversative conjunctions, which overrule the previous clause containing polarized words (e.g., “I like it, but it is not worth it”)
The sentiment score per tweet is then calculated by summing up the polarity scores of the context clusters and dividing it by the square root of the number of words (De Haan, 2020). It ranges from -2 to 2, where a score of -2 is an extremely negative tweet, and 2 indicates an extremely positive tweet. A score of 0 means the sentiment is neutral.
4.3 Statistical models
Since the observations in the final dataset are information on N individual firms observed over multiple T time periods (ten quarters, from 2008 to mid-2010), this data is classified as panel data. The data have both cross-sectional and time-series dimensions. This dataset is balanced because all the individual firms are observed in all time periods (Ti = T). There are three types of models suited for panel data: the pooled model, the fixed effects model, and the random effects model. The pooled model specifies constant coefficients and is the most restrictive panel data model. The fixed effects model allows the individual-specific effects αi to be correlated with the regressors x, in which αi is included as intercepts. Every
individual has a different intercept term and the same slope parameters. Lastly, the RE model assumes that the individual-specific effects αi are distributed independently of the regressors x. In this model, αi is included in the error term, and every individual has the same slope parameters and a composite error term (Econometrics Academy, 2013).
These models will be estimated to examine the extent to which the predictors are able to predict each other and to test the hypotheses H1a - H1e (and H2-H4, on which will be
elaborated in 4.4.6, 4.4.7 and 4.4.8, respectively). The regression models will be estimated by making use of the R package “plm”. For every DV, three different models are estimated which have their own model specification. For estimating pooled models, the model
specification “pooled” is used, for fixed effects “within” and for estimating a random effects model, the specification “random” is utilized. The models included the following
components:
• The associated proxy from the BAV dataset as the DV;
• The score for the SMBBE dimension, that is, the ratio of tweets related to a certain dimension and the net tweet sentiment of the SMBBE dimension to control whether it is the hypothesized score that is related to the self-indicated dimension, or the
36
• The score for the other SMBBE dimensions to check whether the proxy really does cover the dimension;
• A random intercept to control for the brand-specific effects;
• Dummy variables to control for the time-specific effects;
To assess which model is most suitable for this type of data, first, a Breusch-Pagan Lagrange Multiplier test is performed to test whether it is significantly different from zero. It appears that the LM test is significant (p-value <0.001); therefore, the variances across entities are not zero. There are significant differences across units. This implies that a panel effect can be observed (Torres-Reyna, 2007). Based on this, a fixed effects or a random effects model is used. Next, a Hausman test is performed to find out whether the random effects estimator actually is more efficient for this data. It tests whether the unique errors ( are correlated with the regressors (Greene, 2008). By executing the Hausman test, it appears that there is a significant correlation between the two (p-value<0.001). Therefore, decided is to continue with the fixed effects model specification for further analyses.
4.4 Analysis per hypothesis
The analyses used for assessing the hypotheses on the dimensions of SMBBE (brand awareness, brand associations, brand loyalty, perceived quality, and brand relevance), the relationship between these components and future firm performance, and the effects of the moderating variables on this are discussed in the upcoming sections.
4.4.1 Brand awareness (H1a)
The number of tweets posted about a particular brand serves as the measurement of brand awareness. The more tweets posted, the higher the volume of tweets, and thus the higher the brand awareness established for that brand in that period. Therefore, the score for a firm’s brand awareness will be the ratio between the number of tweets on a brand in a certain period and the total amount of tweets of competing brands (in the same category) in the same period. When a brand is receiving a lot of attention in a period, the number of tweets targeted at that firm is relatively high, and this should correlate with the self-indicated firm’s
37
can be assessed whether the share of tweets related to SMBBE’s brand awareness is positively related to self-indicated future brand awareness, as hypothesized in H1a.
4.4.2 Brand associations (H1b)
For this hypothesis, hypothesized is that the higher the share of tweets related to brand associations in a given period, the higher the brand associations is for that brand in the same period. For each brand in every period, interactions are collected, which contain words associated with brand associations. As mentioned in 2.7.2 and according to Keller (1993), there are three major types of brand associations: attributes, benefits, and attitudes. Words associated with these classifications are: “performance”, “appearance”, “experience”, “encounter”, “employees”, “service”, “assistance”. Also, a sentiment analysis is executed to obtain polarity sentiment scores for each interaction, after which the net tweet sentiment is determined. The score for a firm’s brand associations will thus be the share of tweets related to brand associations in one period, relative to the total amount of tweets on that brand for that period, and the net tweet sentiment. These variables are then correlated with the self-indicated brand associations for that brand in the same period. Data on the self-indicated brand
associations are retrieved from the BAV dataset. A proxy for this dimension will be compromised of the variable “esteem”, which consists out of the (weighted) variables “reliable”, “quality”, and “leader” and covers what associations people have with the brand. By executing a correlation test and reviewing the significance of the variables in the
regression model, it can be assessed whether the share of tweets related to SMBBE’s brand associations is positively related to self-indicated future brand associations, as hypothesized in
H1b.
4.4.3 Brand loyalty (H1c)
38
on the self-indicated brand loyalty is retrieved from the BAV dataset. A proxy for this
dimension will be comprised of two variables, that is, “total users” and “total preference”. The former represents the number of people that indicated they occasionally/often buy or use the brand, on a scale from 1 to 100, while the latter represents the number of people that indicated they strongly/moderately consider buying or using the brand in the future, on a scale from 1 to 100. To find out whether the two variables can be merged into one proxy, the reliability is assessed by computing a reliability coefficient. Since Cronbach Alpha’s and especially Pearson correlation are found to be inadequate measures of the reliability of a two-item measure (Cramer, Atwood & Stoner, 2006; O’Brien, Buikstra & Hegney, 2008), a Spearman-Brown test is performed. The Spearman-Spearman-Brown coefficient is less biased and is suggested as the most appropriate reliability statistic (Eisinga, Te Grotenhuis & Pelzer, 2013). It appears that the two variables, “total users” and “total preference” have a strong, positive correlation, which was statistically significant (ρ = 0.78, p-value = <0.001). Therefore, a correlation test is performed on SMBBE’s brand loyalty and the merged variable from the BAV dataset to find out whether the share of tweets related to brand loyalty is positively related to self-indicated future brand loyalty.
4.4.4 Perceived quality (H1d)
39
related to SMBBE’s perceived quality is positively related to self-indicated future brand awareness, as hypothesized in H1d.
4.4.5 Brand relevance (H1e)
To measure the extent to which customers find the brand to be relevant to their needs, interactions are collected, which contain the topic brand relevance and any words associated with it. Words associated with brand relevance are “relevant”, “relevance”, “unique”, “differentiate”, “needs”, “distinct” (Stahl et al., 2012). A sentiment analysis is executed as well to obtain polarity sentiment scores for each interaction, after which the net tweet sentiment is determined. The score for a firm’s brand relevance will be the share of tweets related to brand relevance in one period, relative to the total amount of tweets on that brand for that period, and the net tweet sentiment. These variables are then regressed on the self-indicated brand relevance for that brand in the same period. The dimension brand relevance is also incorporated in the BAV dataset, hence this variable serves as the self-indicated brand relevance. To investigate whether the share of tweets related to SMBBE’s brand relevance is positively related to self-indicated future brand relevance, a correlation test is once again performed.
4.4.6 Tweets on firm performance (H2)
In order to establish the relationship between the predictors and future firm
performance, it is decided to first assess the extent to which information from the BAV and SMBBE data are related to each other, in line with work of De Haan (2020). The correlations between the BAV and SMBBE variables are shown in Table 4. For a predictor to be valuable, it is critical that historical information of one predictor supports explaining future values of that and other predictors. Said differently, it should add new and unique information to the existing knowledge. In the situation that past values of one predictor are strongly correlated to future values of that predictor, it does not attain that goal since it does not add much new information. The same result is yielded if one predictor is fairly well able to predict future values of another predictor, thus the other predictor does also not contain new information.
