Abstract
How can we know what stakeholders think and feel about brands in real time and over time? Most brand reputation measures are at the aggregate level (e.g., the Interbrand “Best Global Brands” list) or rely on customer brand perception surveys on a periodical basis (e.g., the Y&R Brand Asset Valuator). To answer this question, brand reputation measures must capture the voice of the stakeholders (not just ratings on brand attributes), reflect important brand events in real time, and connect to a brand’s financial value to the firm. This article develops a new social media–based brand reputation tracker by mining Twitter comments for the world’s top 100 brands using Rust–Zeithaml–Lemon’s value–brand–relationship framework, on a weekly, monthly, and quarterly basis. The article demonstrates that brand reputation can be monitored in real time and longitudinally, managed by leveraging the reciprocal and virtuous relationships between the drivers, and connected to firm financial performance. The resulting measures are housed in an online longitudinal database and may be accessed by brand reputation researchers.
Keywords
Social media is playing an increasingly important role in enabling conversations about brands (Fossen and Schweidel 2019; Hewett et al. 2016; Kubler, Colicev, and Pauwels 2020). In the social media age, listening online to how brands are talked about is critical for brand management because the comments are from people who care about brands, and they are real-time and dynamic.
Many brand measures currently exist, but they are mostly at the aggregate level, are survey-based, and are typically only available annually. Examples of these are the Interbrand “Best Global Brands” list, the Forbes “The World’s Most Valuable Brands” list, the Kantar Millward Brown “BrandZ Top 100 Global Brands” list, among many others. These brand measures rank leading companies in terms of their overall brand value.
Other brand measures (e.g., the Y&R Brand Asset Valuator) measure multiple dimensions of customer brand perceptions using surveys, but the actionability of the dimensions is limited, given that the measured dimensions (differentiation, relevance, esteem, and knowledge) do not map in a straightforward way to strategic business decisions. Nevertheless, researchers who have attempted to research brand perceptions over time have resorted to such measures (and the Brand Asset Valuator in particular) to conduct their research (e.g., Huang and Dev 2020; Mizik 2014; Mizik and Jacobson 2008; Stahl et al. 2012; Tavassoli, Sorescu, and Chandy 2014).
To provide a brand measure that exploits this new social environment, we develop a real-time longitudinal brand tracker using social media. The tracker provides a new window into what stakeholders think and feel about brands, in their own voice. Such data provide timely, actionable information about how firms should manage brands. The tracker facilitates both longitudinal analysis and exploration of brands in a real-time and granular way.
This new metric is compiled by mining Twitter data for 100 leading global brands for comments on specific drivers and subdrivers on the basis of the customer equity framework (Rust, Zeithaml, and Lemon 2000; Rust, Lemon, and Zeithaml 2004). The database is available to all academic researchers for performing research on the antecedents and consequences of brand reputation changes over time, at a level of granularity not previously available. Table W1 in the Web Appendix compares the proposed brand reputation tracker with existing brand measures to illustrate the gap.
Our brand reputation tracker contributes to brand management by providing actionable data and analytics in a straightforward manner that facilitates strategic brand decisions. Results from our dynamic multivariate panel vector autoregression (VAR) model show that the three drivers have a brand–value reciprocal relationship as well as a brand–relationship–value virtuous circle. A firm thus can leverage either the reciprocal relationship or the virtuous circle, drawing on whichever driver(s) for which the firm has better leverage. We further demonstrate that the three drivers have real-time, short-term, and longer-term impact, collectively and individually, on abnormal stock returns. The findings provide important implications for managing brands on the basis of the dynamics and the tempos of the drivers.
The tracker also builds on the literature connecting social media mining and marketing. Data mining of Twitter feeds and other social media, for example, have been used to measure brand sentiment (Hewett et al. 2016) and other marketing-relevant metrics (e.g., Schweidel and Moe 2014; Tirunillai and Tellis 2012, 2014). By applying our methodology to three social media platforms (Twitter, Facebook, and Instagram) that are distinct in data and nature of interaction, we demonstrate that the three sets of trackers converge—evidence that social media mining can be used to track brand reputation in real time and that social media fluctuations reflect important brand events.
Our work also adds to the literature and management of corporate reputation by bridging brand reputation and corporate reputation and by providing actionable drivers for managing corporate reputation. Brand reputation is similar to corporate reputation when a firm uses a branded-house strategy (e.g., Google), and it is a component of corporate reputation when a firm uses a house-of-brands strategy (e.g., Procter & Gamble). While the corporate reputation literature has established the importance of corporate reputation as a strategic asset (Ferguson, Deephouse, and Ferguson 2000; Fombrun and Shanley 1990; Lange, Lee, and Dai 2011; Rindova et al. 2005) and as a driver of financial performance and firm value (Pfarrer, Pollock, and Rindova 2010; Raithel and Schwaiger 2015), that literature has tended not to focus on marketing actions. We show that the time series of brand reputation, drivers, and subdrivers captures important brand and firm events; for example, the fluctuations of Facebook’s brand and value drivers time series coincide with its unauthorized account licensing scandal in March 2018, while Google’s innovative subdriver time series captures its late 2018 announcement of updating many algorithms. This shows that brand reputation, a key element of, and sometimes equal to, corporate reputation, can be monitored using the tracker and managed by the drivers, which helps drive firm financial performance.
The tracker also contributes to the customer equity literature. To date, most customer equity research has either relied on cross-sectional surveys for data collection (e.g., Gao, Melero-Polo, and Sese 2020; Ou et al. 2014; Ou, Verhoef, and Wiesel 2017; Rust, Lemon, and Zeithaml 2004; Vogel, Evanschitzky, and Ramaseshan 2008) or sacrificed granularity by using aggregate acquisition and retention statistics (e.g., Gupta, Lehmann, and Stuart 2004; Kumar and Shah 2009; Villanueva, Yoo, and Hanssens 2008; Wiesel, Skiera, and Villanueva 2008). We show that the tracker correlates significantly positively with other well-known aggregate brand rankings, such as Interbrand, Forbes, and BrandZ, and correlates significantly positively with YouGov’s daily brand measures of brand word of mouth (WOM) and brand buzz. This demonstrates that the customer equity literature can be renewed with our new social media tracking methodology that links modern social media marketing actions to customer value and facilitates longitudinal analysis and exploration of customer equity drivers in a more granular way. We also broaden the conceptual nature of customer equity to include all stakeholders, rather than just current and future customers of the brand.
In the following sections, we first conceptualize brand reputation drawing on multiple conceptual sources. Second, we develop our social media tracking method to track and monitor brand reputation. Third, we present empirical evidence based on 130 weeks’ tracking data that show that this brand reputation tracker can be used to monitor and manage brand reputation and competition dynamics and is accountable for a firm’s abnormal stock returns. Fourth, we generalize the tracker to multiple social media platforms and validate the tracker using other brand measures. Finally, we discuss implications for brand managers and researchers.
Conceptualizing Brand Reputation
Brand Reputation
We define brand reputation as the overall impression of how stakeholders think, feel, and talk about a brand. This is typically due to brand events that affect firm financial performance. This definition has the following characteristics: (1) it is about all stakeholders (current and potential customers, employees, partners, and investors), not just the current or potential customers; (2) it has thinking, feeling, and talking components (not just knowledge about brands); (3) it can reflect actual brand events (e.g., controllable marketing activities, uncontrollable public events); and (4) it connects to firm financial performance.
We highlight “stakeholders” to indicate a broader view of customers, which is an integration of the corporate reputation and the marketing literature. This view is customer-centric, but is broader, considering all stakeholders as customers, including current (current and churning current customers), potential (competitors’ customers and noncustomers), internal (employees), and external (investors and partners) customers. For example, Berry (2000) considers both external customers and internal employees as relevant for building service brand equity. Hanssens, Rust, and Srivastava (2009) similarly urge the need to take a broader view of the stakeholders of marketing strategy by including the investor community as a customer. Nguyen, Calantone, and Krishnan (2020) demonstrate that social media emotional WOM influences investors’ decisions on holding a firm’s stocks. This broader view of customers reflects that brand reputation can be perceived by nonrelationship brand stakeholders who can influence the brand’s financial performance.
Brand reputation should be based on whatever stakeholders say about a brand, meaning what is explicitly expressed about their thinking and feeling, not what is implicitly inferred. Stakeholders on social media can discuss anything about a brand. It can be brand experience, opinions about brand events, or simply personal sentiments about a brand. It can be positive, neutral, or negative in varying degrees. The overall impression of a brand may be summarized by what stakeholders say about a brand on social media (Hewett et al. 2016).
Changes in brand reputation typically result from actual brand events. These brand events can be controllable marketing actions and activities as well as uncontrollable public events about brands. This characteristic emphasizes the actionability of brand reputation, allowing marketers to actively manage a brand’s reputation and to track the reputation for risk and crisis management. Such actionability is the focus of models of return on marketing (Rust, Lemon, and Zeithaml 2004).
Brand reputation should be value relevant, that is, connecting to a firm’s financial performance. This value relevance reflects investors’ expectations about the financial value of current, potential, internal, and external customers to the firm. Value-relevant brand reputation thus is a corporate asset and a driver of firm financial performance, as defined in the management literature (e.g., Ferguson, Deephouse, and Ferguson 2000).
Next, we compare brand reputation with other related concepts in the management and the marketing literature streams to illustrate the conceptual nature of brand reputation. Figure 1 illustrates the relationship between brand reputation and other related concepts. It shows that our brand reputation concept lies in the intersection between the concepts of corporate reputation, brand equity, and customer equity. It has consequences for all of the firm’s stakeholders (not just customers), focuses on brand thinking and feeling, and emphasizes marketing actions to drive firm value.
Conceptual sources of brand reputation.
Corporate Reputation
In the management literature, corporate reputation is generally defined as an overall appraisal of a company by its stakeholders, which is the result of the company’s past actions and predictions about the company’s future (Ferguson, Deephouse, and Ferguson 2000). Such an overall appraisal can be thinking and/or feeling based, pertaining to, for example, the company’s quality or capability (Boivie, Graffin, and Gentry 2015), people’s admiration of the company (Dowling 2016), or people’s knowledge and emotions about the company (Hall 1992). With its broad scope, taking all stakeholders into consideration (Argenti and Druckenmiller 2004), and with the appraisal being at the corporate level, corporate reputation has a broader and higher-level focus on components such as leadership, social responsibility, product and service, workplace/employee, and management/financial performance as well as their consequences on firm performance (Parker, Krause, and Devers 2019; Raithel and Schwaiger 2015; Roberts and Dowling 2002).
Compared with corporate reputation, which is the overall appraisal of a firm held by stakeholders that has consequences for firm performance, brand reputation is the analogue for companies that use a branded-house strategy (e.g., FedEx, Google, Apple). In addition, it is a component of corporate reputation when the company uses a house-of-brands strategy (i.e., having multiple brands to represent the company; e.g., Procter & Gamble).
Brand Equity
The brand equity literature emphasizes customers’ overall impression about a brand, even though there is disagreement on the scope of the impression. Some take a broader view. Aaker’s (1995) brand equity definition includes mainly brand loyalty, brand awareness, perceived quality, and brand associations, constituting various assets and liabilities associated with a brand and value derived from them. Keller’s (1993) customer-based brand equity concept defines brand equity in terms of the individual consumer’s brand knowledge, which influences the consumer’s reaction to the brand’s marketing mix. Brand knowledge is broad, including all knowledge about a brand that the consumer has, such as higher-level brand awareness and brand image, and all lower-levels associations, such as brand recall and brand associations. Alternatively, some take a narrower view. Rust, Zeithaml, and Lemon (2000) and Rust, Lemon, and Zeithaml (2004) define brand equity as customers’ subjective or emotional appraisal of a brand, above and beyond its objectively perceived value. This inconsistency in the scope of brand equity motivates some studies trying to bridge brand equity with customer equity (e.g., Gani and Grobler 2019; Leone et al. 2006).
Brand reputation reflects both the knowledge and emotions held by all stakeholders about a brand. Our broader view aims to capture how a broader set of stakeholders thinks, feels, and talks about a brand, not limited to whether they are current or potential customers.
Customer Equity
Reflecting a customer lifetime value view, the existing customer equity literature focuses on the contribution of a brand’s single stakeholder (i.e., customers) to a firm. Rust, Lemon, and Zeithaml (2004, p. 110) define customer equity as the total of the discounted lifetime values summed over all of the firm’s current and potential customers. Blattberg and Deighton (1996) define customer equity as the discounted profit stream and explore customer equity in terms of the optimal trade-off between acquiring and retaining customers. Gupta, Lehmann, and Stuart (2004) similarly define the value of the customer base as the expected sum of discounted future earnings.
To enhance customer value, the existing studies focus on either customer equity drivers (mostly perceptual) or marketing investments. For customer equity drivers, research has shown that the three drivers in the customer equity framework—value, brand, and relationship (Rust, Zeithaml, and Lemon 2000)—can improve customer value, as captured by loyalty intentions (Vogel, Evanschitzky, and Ramaseshan 2008), customer loyalty (Ou et al. 2014; Ou, Verhoef, and Wiesel 2017), and customer experience quality (Go, Melero-Polo, and Sese 2020). For marketing investments, the research stream on the marketing–finance interface further links customer equity to shareholder value (Gupta, Lehmann, and Stuart 2004; Kumar and Shah 2009; Schulze, Skiera, and Wiesel 2012; Skiera, Bermes, and Horn 2011; Wiesel, Skiera, and Villanueva 2008).
Brand reputation is similar to customer equity in that both can be driven by the customer equity drivers and linked to firm financial performance. The focus is on what firms do, in terms of customer equity drivers or marketing investments, to influence firm financial performance.
The Social Media Tracking Method
We use a multistage, theory-data iterative process to develop the tracker, detailed next and shown in Figure 2. We briefly summarize these steps and explain the logic behind them, and describe this process in detail in the next sections.
A multistage, theory-data iterative process to develop the tracker.
Theory-driven text mining. Use the customer equity framework to establish a three-driver tracker. This makes the brand reputation outcome theory driven and thus explainable.
Managerially actionable subdrivers. Develop the subdrivers of brand reputation on the basis of managerial actionability. This makes the drivers managerially relevant and actionable.
Customized dictionaries. Generate dictionaries for the subdrivers using real-world stakeholders’ own words. This extracts what they think and feel about a brand in conversation and in context.
Real-time brand reputation tracking. Collect data in real time and from multiple social media platforms. This enables firms to respond quickly and allows the methodology to be applied across platforms.
Leverageable driver synergies. Provide evidence for the dynamics and tempos of the three drivers. This provides managerial and theoretical insight as to the internal relationships between the drivers and how a firm can leverage the synergies of the three drivers.
Financially accountable brand reputation. Establish the accountability of the tracker for firm abnormal returns. This makes brand reputation and its drivers financially accountable, rather than just time-series of brand reputation fluctuations.
Establish a Three-Driver Tracker
We employ the driver structure from the customer equity framework (Rust, Lemon, and Zeithaml 2004; Rust, Zeithaml, and Lemon 2000) to develop our tracker. This framework organizes the factors driving customer lifetime value and contribution to the firm into three main drivers, which themselves may be broken down further into subdrivers. Value equity is the rational and objective aspects of a brand, such as quality and price. Brand equity is the subjective feeling that a customer has about the brand, such as brand sentiment and brand image. Relationship equity is the ties between the customer and the brand, above and beyond the value equity and brand equity, such as brand community building and personal connection.
The following considerations contribute to the choice of this framework: First, the conceptual attraction of this framework has been well recognized in the academic community and recognized with several article and book awards. Second, the three customer equity drivers have been validated conceptually and empirically in many subsequent studies, using data from multiple countries; incorporating both perceptual survey and behavioral data; and considering industries, firms, and consumer characteristics; and they have been gauged using various firm performance variables (e.g., loyalty intentions, future sales, customer experience quality) (Gao, Melero-Polo, and Sese 2020; Leone et al. 2006; Ou et al. 2014; Ou, Verhoef, and Wiesel 2017; Vogel, Evanschitzky, and Ramaseshan 2008). Third, this framework was designed to map to strategic expenditures and thus has high managerial actionability. The drivers and subdrivers have been shown to link to return on marketing, an important characteristic that helps connect brand reputation to firm financial performance. Its managerial relevance is reflected in this framework being applied at many leading companies worldwide. Fourth, the value and brand drivers together capture the thinking and feeling aspects of brand reputation. It is the consensus that a brand metric should include both aspects (Huang and Dev 2020; Lovett, Peres, and Shachar 2013; Vogel, Evanschitzky, and Ramaseshan 2008).
Choosing the Social Media
Many different social media platforms are used to discuss brands (e.g., Twitter, Facebook, Instagram). To construct a dynamic tracker of sentiment about brands on social media, we chose Twitter for the following considerations: (1) most Twitter accounts are public, meaning that conversations on Twitter have a larger impact on public perception of the brand, whereas many other social media platforms (e.g., Facebook) default to private communications; (2) most brands maintain an active presence on Twitter, which means that brand conversations are continuously updated and are available for public access; and (3) Twitter provides a publicly available application programming interface that can identify conversations about the brands, for example, using username “@coach,” rather than “coach” to identify conversations about the brand ensures precision (which is the number of relevant tweets retrieved divided by the number of all tweets).
Choosing Brands to Monitor
The choice of brands to be monitored is based on various prominent industry rankings on brands. These rankings included Forbes’s World's Most Valuable Brands, BrandZ’s Top 100 Most Valuable Brands, Interbrand’s Best Global Brands, CoreBrand’s Top 100 Brand Power Rankings, Credit Suisse Research Institute’s Great Brands of Tomorrow, Ad Age’s Social Media Brand Ranking Top 10, UTA Brand Studio’s Brand Dependence Index, and Reputation Institute’s Global Reputation Pulse U.S. Top 15.
Once all the brands are tabulated, any brand that appears twice or more across the lists is added to our database as a brand to be included in the tracker. Table W2 in the Web Appendix lists the brands included in the tracker. The database consists of 100 global brands across a broad range of industries such as manufacturing, wholesale trade, retail trade, transportation and warehousing, information, finance and insurance, professional, scientific, and technical services, and accommodation and food services. Both internet brands and traditional brands, and both corporate brands and individual brands, are included.
Collecting Tweets Using Twitter Username
After the list of brands is established, we identify the top Twitter username (i.e., Twitter handle) associated with each brand. In deciding which handle to use, we apply the rule of the “top returned handle,” that is, the brand handle (ignoring subhandles or regional handles) returned in the top search result by searching the brand name on Twitter. If no corporate brand is found, we check @brandname to see if it is a valid handle and double-check if users mention @brandname tweets at least once in the last month. The technical details of the Twitter data collection are shown in the Web Appendix, and Table W2 in the Web Appendix lists the brand names and handles used for collecting the data.
Table 1 summarizes the final 11 subdrivers for the three brand reputation drivers, including their conceptual descriptions and the final positive and negative dictionaries used in the data collection. Specifically, they are (1) value driver (price, service quality, and goods quality), (2) brand driver (cool, exciting, innovative, and social responsibility), and (3) relationship driver (community, friendly, personal relationships, and trustworthy).
Brand Reputation Drivers, Subdrivers, Descriptions, and Dictionaries.
Notes: The goods quality subdriver applies to goods brands only, while the service quality subdriver applies to all brands.
The 11 subdrivers and their dictionaries are theoretically derived and empirically validated by multiple rounds of data collection and evaluation. They capture nicely the social media language and technology, while preserving the conceptual nature of the three brand reputation drivers as laid out in Rust, Lemon, and Zeithaml’s (2004) framework. For example, for the value driver, they have quality and price as the subdrivers, and we further refine the quality subdriver into service quality and goods quality, a reflection of the service economy and the distinctiveness of service quality from goods quality. The dictionaries of the three subdrivers contain keywords that are unique in a social media setting and use stakeholders’ daily language, such as “joy” for the price subdriver, “lazi” for the service quality subdriver, and “beauty” for the goods quality subdriver.
For the brand driver, Rust, Lemon, and Zeithaml (2004) have corporate citizenship and ethical standards as subdrivers, and we have social responsibility as one of the subdrivers. Our brand driver is the most social media–centric driver, with three of the four subdrivers reflecting stakeholders’ usage of social media language in expressing their thinking and feeling about brands, such as “cool,” “exciting,” and “innovative.” The dictionaries of the subdrivers also reveal the language stakeholders use, such as “sexi” for the cool subdriver, “thrill” for the exciting subdriver, “intellig” for the innovative subdriver, and “give” for the social responsibility subdriver.
For the relationship driver, Rust, Lemon, and Zeithaml (2004) include community as one of the subdrivers, and we add friendly, personal relationships, and trustworthy to capture that the new information and communication technologies connect stakeholders more closely to companies and their brands. These subdrivers include both the interaction and communication process of a relationship as well as the trustworthy outcome of a relationship. The dictionaries of the subdrivers are unique in suggesting new terms when communicating with stakeholders, such as “famili” for the community subdriver, “open” for the friendly subdriver, “intim” for the personal relationships subdriver, and “transpar” for the trustworthy subdriver.
Empirical Evidence
The data set covers the week of July 1, 2016, to the week of December 31, 2018, 130 weeks in total. 1 The brand panel data contain 13,000 brand-week observations of 100 unique brands. 2 We measure volume and sentiment on the three drivers and subdrivers. Table 2 presents the mean, standardization deviation, minimum, maximum, and correlations among the overall brand reputation, drivers, and subdrivers. All pairs of correlations are significant at the .001 level but are moderate in effect, indicating a balance between representativeness and uniqueness.
Descriptive Statistics: Median, Minimum, Maximum, and Correlations Among Brand Drivers and Subdrivers.
Notes: The unit of analysis is brand-week. Variables 1–11 are subdrivers, variables 12–14 are drivers, and variable 15 is the overall brand reputation. All variables have 13,000 observations, except the goods quality subdriver, which has 8,580 observations. All correlation coefficients are significant at .000. Mean, SD, Min, and Max are calculated using raw scores, and correlations are calculated using the normalized scores, shown in Web Appendix Equation W4. Negative minimums indicate the number of negative keyword mentions is greater than the number of positive keyword mentions.
Tracking Brand Reputation and Competition
Monitoring brand engagement
Brand tweet volume can be viewed as how engaged stakeholders are with a brand. The more discussion a brand can generate, the more engaged it is with its stakeholders. Volume of social media discussion (e.g., tweets and retweets on Twitter) is considered to capture social media engagement (Colicev et al. 2018). It has also been shown to impact brand financial results (Kumar et al. 2013).
In our data set, the average number of tweets collected per week per brand was 14,102 (SD = 46,310, min = 0, max = 1,660,963), which is substantial, indicating that the brand engagement on social media is high. This varies highly by brand, however. For example, the brands with the largest mean weekly tweet volume were Amazon (mean = 1,660,963, the week of February 23, 2018), T-Mobile (mean = 1,180,385, the week of November 18, 2016), and Google (mean = 824,432, the week of February 23, 2018). 3 From the Black Friday and post-Thanksgiving weekend of November 18–20, 2016, T-Mobile rolled out a Magenta Friday promotion, offering two additional lines free to both existing and new customers. T-Mobile has 1.4 million Twitter followers, and this promotion no doubt generated hot discussion. By comparison, examples of brands with the smallest mean weekly tweet volumes were HSBC, Kraft-Heinz, and Canon, the latter of which has zero tweets. This variation is not surprising, given that the decisions to tweet about a given brand will be driven by various factors, including some that are related to the brand, but also factors that are person-related or intrinsic to the individual (e.g., Toubia and Stephen 2013).
Table W3 in the Web Appendix shows that among the three drivers across all brands, in terms of average volume, the brand driver (N = 387) has a higher volume than the value driver (N = 303) and the relationship driver (N = 165). This pattern in engagement volume among the three drivers may imply that the brand driver captures brand events more quickly and closely.
Monitoring brand sentiment
In Table W3 in the Web Appendix, we show the frequencies of net, positive, and negative words, as defined in our dictionaries, and calculate the positive to negative ratio for brand reputation, drivers, and subdrivers. The higher the ratio, the more positive sentiment of the reputation, drivers, and subdrivers. We can see that, in terms of sentiment ratio (proportion of positive to negative volume), the general sentiment is positive (all greater than 1). For brand reputation, the brands in the tracker, in general, have positive sentiment (3.91). This is understandable because they are top brands worldwide. For drivers, the brand driver is the most positive (13.09), followed by the relationship driver (5.58), with the value driver having the least positive sentiment (2.34). For subdrivers, the top three positive subdrivers are exciting (20.28), cool (15.00), and innovative (14.88), whereas the least positive subdrivers are price (1.45), friendly (3.65), and community (4.36).
For subdrivers of the value driver, price (1.45) is relatively ambivalent, and service quality (7.35) and goods quality (5.23) are both quite positive. This indicates that people talk about price for both positive and negative reasons but mainly talk about quality when it is positive.
For subdrivers of the brand driver, the sentiment of all subdrivers is generally positive; especially for exciting, which has the highest ratio (20.28) among all subdrivers, indicating that it is a powerful driving force for brand reputation. The second most influential subdriver is innovative (14.88). It is the most talked-about subdriver and is highly positive.
For subdrivers of the relationship driver, the most positive subdriver is personal relationships (12.38), suggesting the importance of establishing a personal relationship with stakeholders. When people talk about this, it is overwhelmingly positive.
Monitoring competition
The brand reputation tracker can be used for monitoring brand fluctuations and tracking competitive dynamics at the brand reputation, driver, and subdriver levels. We illustrate this use using two competitive dyads: two technology service brands, Facebook and Google, and two technology goods brands, Apple and Samsung, for 2018. Results and discussions of the Apple and Samsung dyad are presented in the Web Appendix and Web Appendix Figures W1, W2a–W2c, and W3a–W3i.
Figure 3 shows the time series of their brand reputation (blue line for Facebook and red line for Google), Panels A–C of Figure 4 show the time series of the three drivers, and Panels A−D of Figure 5 show the time series of selected subdrivers.
Brand reputation: Facebook versus Google 2018.
Time series for the three drivers.
Time series for the selected subdrivers.
The brand reputation time series show that Facebook appears to have higher brand reputation than Google for the first three quarters of 2018. In the fourth quarter, the gap decreases.
Facebook’s brand (Figure 4, Panel A, blue line) and value (Figure 4, Panel B, blue line) drivers time series show that there was a negative spike from the week of March 19, with the brand driver scores plummeting from .575 to .045 to −.218 in two weeks. This coincided with the revelation of the scandal that Facebook permitted the unauthorized licensing of 30 million people’s accounts to Cambridge Analytica, a data firm used by Donald Trump’s 2016 presidential campaign to target voters. From the week of September 14 to the end of the year, Facebook’s brand driver scores reached a long depression with an average of −.332, compared with an average of .029 the month before it. This corresponded with the event that occurred in September, when 50 million Facebook accounts and sensitive personal data were hijacked. The larger scale and more severe data leakage of this negative event are reflected in the more enduring plunge of the brand driver, indicating that stakeholders were quite concerned about the consequences of this significant personal data hijacking. According to Facebook, the company saw that unusual activities began on September 14, and on September 28 the news came out. The brand reputation tracker captures this negative event in real time, as well as its carryover effect.
The relationship driver time series of Google (Figure 4, Panel C, red line) reveals that Google in general underperforms Facebook in this driver. This is understandable because Facebook is a social networking platform for people to establish and maintain their relationships. Nevertheless, in the last week of September and the first week of October, Google had its second-highest relationship driver score (.240) of the year, surpassing Facebook in this driver. September 27th was Google’s 20th anniversary, and the company also updated many algorithms of the important services it provides.
We further investigate which subdrivers are most accountable for the ups and downs of the brand reputation, shown in Figure 5, Panels A−D. Panel A shows that Facebook outperformed Google in the exciting subdriver, whereas Panel B shows that Google did a better job in the innovative subdriver, especially in the fourth quarter of 2018, when it updated many algorithms. Both subdrivers capture stakeholders’ differential perceptions about the two brands.
Panel C shows that Facebook mainly bested Google in the personal relationships subdriver, but the brand is not considered more trustworthy than Google, as shown in Panel D. This indicates that the trustworthy subdriver can be an opportunity for Google to capitalize on but should be a pain point for Facebook to deal with, especially with all the data leakage negative events.
Managing the Dynamics of Brand Reputation Drivers
We have shown that the tracker reflects important brand events and can be used to monitor competition. In this subsection, we further examine the internal relationship of the three drivers for managing brand reputation.
Dynamic multivariate VAR model
We use a rigorous dynamic multivariate VAR model, estimated with generalized method of moments, to simultaneously estimate the three drivers as a system of equations (Abrigo and Love 2015; Love and Ziccino 2006). The estimator is dynamic, as the current realization of the endogenous variables (i.e., valuet, brandt, and relationshipt) is influenced by their past values (i.e., valuet − 1, brandt − 1, and relationshipt − 1). The inclusion of the lagged-one-period endogenous variables considers the cumulative effect of the drivers over time.
In the model, the predictors include the lagged-one-period values of the three endogenous drivers (
where i = Brand (there are 100 brands), t = Week (there are 130 data weeks),
First, we examine whether each of the three drivers is stationary using a Fisher-type test (Choi 2001). The test has the null hypothesis that all the brand time series contain a unit root. It assumes the data are generated by a first-order autoregressive process; thus, we specify an augmented Dickey–Fuller unit root test on each brand with one lag of the first-differenced driver to remove the higher-order autoregressive components of the series. To mitigate the impact of cross-sectional dependence, we also follow Levin, Lin, and Chu’s (2002) suggestion to demean the data. All test statistics for the three drivers, respectively, are significant at the .001 level, which rejects the null hypothesis of having a unit root. Thus, the test results support that the three drivers are stationary.
Second, we carried out model selection tests to determine the optimal lag order for the model. The results suggest that the first-order panel VAR minimizes the modified Bayesian information criterion (MBIC = −226.449), the modified Akaike information criterion (MAIC = −26.400), and the Quinn information criterion (MQIC = −93.451), compared with the second-order (MBIC = −158.123; MAIC = −24.758; MQIC = −69.459) and the third-order (MBIC = −78.305; MAIC = −11.622; MQIC = −33.972) models. This is expected, given that we have weekly data and discussion about a brand on Twitter changes rapidly and frequently.
Third, we estimated the first-order panel VAR using generalized method of moments–style instruments as in Holtz-Eakin, Newey, and Rosen (1988). Table 3 presents the results. We find that the value driver is influenced by its own lagged value (.214, p = .000), the lagged brand driver (.111, p = .000), and the lagged relationship driver (.122, p = .000). The brand driver is influenced by its own lagged value (.289, p = .000) and is marginally influenced by the lagged value driver (.030, p = .063). The relationship driver is influenced by its own lagged value (.286, p = .000) and the brand driver (.068, p = .013). Hansen’s J-statistic is near zero, confirming that the model is not overidentified.
The Mutual Impacts of Brand Reputation Drivers: Multivariate Dynamic Panel Model.
† p < .1.
*p < .05.
**p < .01.
***p < .001.
Notes: N.A. = not applicable. t denotes the current value, and t − 1 denotes the lagged-one-week value of the respective variables. Instruments include all variables in the equation.
The Granger causality tests 4 confirm that the value driver marginally Granger-causes the brand driver (χ2 = 3.461, p = .063), the brand driver Granger-causes the value driver (χ2 = 11.951, p = .001) and the relationship driver (χ2 = 6.144, p = .013), and the relationship driver Granger-causes the value driver (χ2 = 13.560, p = .000).
We calculate the impulse response function (IRF) confidence intervals using 200 Monte Carlo draws based on the estimated model. Figures W4a–W4d show the relevant IRF figures. The shaded area is a 95% confidence band. The IRF figures show that, in general, the interdriver effects level off in about five to six weeks.
For the impact of the brand driver on the value driver, a shock on the brand driver creates a short-term (lagged-one-week) surge on the value driver, and this surge gradually levels off in five weeks (Figure W4a). For the impact of the value driver on the brand driver, a shock on the value driver has a real-time positive impact on the brand driver. Although it levels off quickly (Figure W4b), its full effect dissipates gradually over four to five weeks.
For the impact of the relationship driver on the value driver, a shock on the relationship driver has a short-term positive impact on the value driver (Figure W4c). Its impact is smaller than the impact of the brand driver (Figure W4a) but is about equally persistent.
For the impact of the brand driver on the relationship driver, a shock on the brand driver has a real-time positive impact on the relationship driver, which levels off more slowly than the impact of the value driver on the brand driver and persists for a longer time period, for about five to six weeks (Figure W4d).
Dynamics of the three drivers
The customer equity framework considers that the three drivers together constitute the bonds that hold the customer to the brand, but it does not specify how the three drivers causally relate to each other. Our empirical examination thus provides original empirical evidence regarding the dynamics of the three drivers. The three rectangular boxes and the white arrows in Figure 6 illustrate the mutual impacts of the drivers. The outer gray arrows depict the financial impact of the three drivers, which we discuss in the next subsection.
The mutual impacts of brand reputation drivers, and their synergies on abnormal stock returns.
Two relationships emerge. First, we find a reciprocal relationship between the brand and the value drivers, with the impact from the brand driver to the value driver being stronger than the reverse. Second, we find a virtuous circle among the three drivers, from the brand driver to the relationship driver, from the relationship driver to the value driver, and finally from the value driver back to the brand driver. The IRF figures further reveal different tempos of the carryover effects among the three drivers.
Together, the two relationships expand our knowledge of how the customer equity drivers relate to each other over time and provide rich implications for managing brand reputation. We discuss their managerial implications in the “Discussion” section.
Establishing the Financial Accountability of the Tracker
Previously, we showed that the time series of the brand reputation tracker can capture important brand events. Next, we further demonstrate that the unanticipated components of the tracker provide additional information to stakeholders about the firm’s abnormal stock returns.
To do so, we match the brand tracker data with the firm’s financial data from the Center for Research in Security Prices (CRSP). After the matching, we obtain 8,710 firm-week observations with 67 single-brand firms that trade in the U.S. stock market. In this matching, individual brands from firms that follow a house-of-brands strategy (i.e., one firm has multiple brands) are excluded (e.g., Pampers) for consistency. Table W4 in the Web Appendix summarizes the industry characteristics of the brands in the tracker, based on their two-digit North American Industry Classification System codes. It shows that we have 40.30% (N = 27) manufacturing brands and 59.70% (N = 40) service brands. Both the manufacturing and the service brands are dominated by technology and information brands, reflecting the nature of the modern information economy.
In calculating abnormal returns, we first estimate a firm’s expected stock returns using the Carhart four risk factors (Carhart 1997; Fama and French 1993) to adjust stock returns for the risk factors and to demonstrate that the drivers provide additional explanations for abnormal returns. Our calculation is similar to that of Nam and Kannan (2014) and Srinivasan et al. (2009), as shown in Equation 2:
where Rit is firm i’s actual stock return in week t, and Rft, t is the risk-free rate of return in week t. We obtain daily stock returns data from the CRSP and collapse them into weekly stock returns to match our weekly brand data. The three Fama–French factors—the risk-free market return rate (Rmt – Rrf, t), the return difference between small-firm and big-firm stocks (SMBt), and the return difference between high and low book-to-market stocks (HMLt)—are accessed from Fama and French’s data library. The momentum factor (UMDt) is the return difference between portfolios of past winners and losers (Fama and French 1993).
We estimate the impact of the unanticipated component of the drivers on the firm’s abnormal return using Equation 3:
where i = Brand (1 to 67; i.e., single-brand firms), t = Week (there are 130 data weeks), ARit = Abnormal return for firm i in week t (ARit = [Rit – Rft, t] – Eretit, where expected return, Eretit, is the predicted value of Rit − Rft, t in Equation 2), UΔX = Unanticipated component of the drivers, d = Driver (1 to 3; i.e., brand, value, and relationship drivers), s = No. of week lag (1 to 2), g = Index for the other brand in the two-way interaction, t-s = Index for time with lag(s) (i.e., the current, lag one week, or lag two weeks),
βd, t − s, γdg, t − s, θt − s = Parameters to be estimated
UΔX, the unanticipated component of driver is the standardized residual estimated by a fixed-effect panel model for each driver, using its lagged-one-week value as the predictor. We include lagged-one-week and two-week drivers as predictors to capture the immediate, short-delayed, and longer-delayed effects of the drivers. 5 Industry sector and year dummy are included as control variables. Industry sector is the set of industry dummies, ranging from 1 to 5 (the manufacturing sector as the baseline). Year dummy is the set of year dummies, ranging from 2016 to 2018 (year 2016 as the baseline). 6 Equation 3 is estimated using a feasible generalized least squares panel model, specifying a heteroskedastic error structure and panel-specific autocorrelation. This allows for flexible autocorrelation across brands and a brand-specific first-order autoregressive process for the error in each brand.
Accountability of drivers
The estimation of Equation 2 shows that the market risk factor (Rmt − Rft, t) has a significant positive effect on stock returns (.801, p = .000), whereas the SMB factor has a significant negative effect (−.001, p = .000). The other two factors are not significant. The positive coefficient for the market risk factor shows that each firm’s stock returns covary with the risk-free market returns, and the negative coefficient for the SMB factor indicates that big firms have higher returns than small firms. The constant is insignificant (−.000, p = .857), consistent with the efficient-market hypothesis.
We then estimate Equation 3 to check the accountability of the residuals of the drivers for a firm’s abnormal return. Table 4 presents the results for the main-effect model and the interaction model, respectively. For the main-effect model, the residual of the brand driver has a real-time positive impact on abnormal returns (.001, p = .034), the residual of the value driver has a short-term positive impact (.001, p = .085), and the residual of the relationship driver has real-time (−.002, p = .028) and short-term (−.002, p = .001) negative impacts but a longer-term positive impact on abnormal returns (.001, p = .079). The information sector has higher abnormal returns than other sectors (.001, p = .013). We find no significant year effect.
Accountability of the Unanticipated Component of Brand Reputation Drivers for Abnormal Stock Returns.
*p < .1.
**p < .05.
***p < .01.
Notes: t denotes the current week value, t − 1 denotes the lagged-one-week value, and t − 2 denotes the lagged-two-week value of the respective variables. UΔ denotes the unanticipated component of brand drivers, estimated as the standardized residual using the lagged-one-week value of the respective variable (i.e., t − 1) as the predictor in a fixed-effect panel model.
Results from the interaction-effect model show that the residual of the brand × value interaction has a real-time positive impact on abnormal returns (.001, p = .089), the residual of the value × relationship interaction has a short-term positive impact on abnormal returns (.001, p = .062), but the residual of the brand × relationship interaction has a real-time negative impact on abnormal returns (−.001, p = .052).
Accountability of driver sentiment
We then take the sentiment of the drivers into consideration by estimating Equation 3, but we separate the residual of the positive and negative sentiments of the drivers into two models.
For the negative sentiment model, we find that the residual of the negative value driver has a short-term negative effect (−.002, p = .041), the residual of the negative relationship driver has a longer-term negative effect (−.003, p = .006), and the residual of the negative brand driver has no impact. The residual of the negative brand × value interaction has a longer-term negative impact (−.001, p = .041), and the residual of the negative relationship × value interaction has a real-time negative impact (−.001, p = .095).
For the positive sentiment model, we find that the residual of the positive value driver has a short-term positive effect (.002, p = .025), the residual of the positive relationship has a short-term negative effect (−.004, p = .000), and the residual of the positive brand driver has no impact. The residual of the positive brand × relationship interaction has a negative effect (−.002, p = .051), but the residual of the positive brand × value × relationship interaction has a marginal positive effect (.000, p = .108).
Together, the results from the two sentiment models show that the negative sentiment of drivers matters more, and more consistently. Furthermore, the marginal positive impact of the three-way interaction from the positive sentiment model confirms the brand–relationship–value virtuous relationship between the drivers found in the VAR model, indicating that this virtuous relationship can be accountable for a firm’s abnormal returns.
Summary
The analysis demonstrates that the three drivers provide additional information for a firm’s risk-adjusted abnormal stock returns in real time, the short term, and the longer term. The results show that the brand driver has a real-time impact and is the dominant driver for abnormal returns, the value driver has a short-term impact and synergizes with the other two drivers, and the relationship driver has a longer-term impact and its positive sentiment synergizes with the other two drivers. This pattern of impact is consistent with the dynamics of the three drivers observed in the previous section.
The impact of the brand driver is more real-time, reflecting that the driver captures stakeholders’ immediate sentiments to brand events or activities, as demonstrated by Hewett et al.’s finding (2016) that online WOM echoes fast and wide in an “echoverse” of the brand’s communication.
The impact of the value driver is more short-term, reflecting that quality and cost do not fluctuate as frequently as brand feelings. This driver reflects the knowledge aspect of a brand, which, according to the brand equity literature (e.g., Keller 1993), can be expected to be more stable than emotional reactions to brand events. Its positive and negative sentiments provide separate information for a firm’s abnormal returns, indicating the need for monitoring both the positive and negative discussions about a brand. Its synergy with the brand driver also indicates that the objective aspects of a brand’s reputation (e.g., price, quality) need to be associated with positive feelings with the brand (e.g., cool, exciting) to benefit a firm’s abnormal returns.
The impact of the relationship driver is longer-term and hinges more on the positive sentiment, reflecting that relationships take time to play out, but the stock market may be myopic with respect to longer-term marketing impacts (e.g., Huang and Trusov 2020; Mizik and Jacobson 2007). Once a good reputation on this driver is built, it benefits a firm’s abnormal returns in the long run, as shown in the longer-term brand–relationship–value synergy for the positive sentiment of the drivers.
Together, the results suggest that the residual of the drivers provides information value for a firm’s abnormal returns, immediately or in a delayed manner, individually and collectively. The dark curved arrows in Figure 6 summarize the analysis. Thus, by monitoring the fluctuations of the drivers, stakeholders can have a more accurate picture about a firm’s financial performance.
Validation
We validate the brand reputation tracker using three approaches. First, we replicate our methodology using two additional social media platforms, Facebook and Instagram, each of which has idiosyncratic features: Facebook focuses on social networking, and Instagram focuses on photo sharing. Second, we establish a nomological relationship with the survey-based YouGov brand data, from which we demonstrate that the tracker is related to YouGov’s brand WOM and brand buzz, and leads to YouGov’s purchase intention. Third, we demonstrate that the tracker correlates significantly positively with three aggregate annual brand measures, Interbrand, Forbes, and BrandZ, showing that the tracker not only converges with the aggregate annual measures but also provides more granular information (both in terms of time interval and drivers) for brand reputation.
Replicating the Tracker Using Other Social Media Platforms
Data
We collected data from Facebook and Instagram, from January 1 to June 30, 2018 (i.e., 26 weeks), for the seven internet brands in the tracker. The data were collected using Crimson Hexagon, with a method similar to what was described in the “Social Media Tracking Method” section. However, the data were more difficult and problematic to collect and analyze, because many brand posts on Facebook are not publicly available, and Instagram concentrates on visual data.
We applied the same dictionaries of the subdrivers to the two social media platforms. For Facebook, we focused on post contents that are available on the firm’s own brand pages. For example, for the Amazon brand, its Facebook page is https://www.facebook.com/Amazon, and a sample post content is “Thank you Amazon for excellent customer service and speedy response! Keep up the great work!” For Instagram, we collected captions and comments on photos that mention the brand handles. For example, the Amazon brand’s Instagram handle is @amazon, and a sample post is “@amazon…I ordered a waffle iron two days ago and they delivered it to this tiny island 30 miles out to sea so quickly.”
In calculating the driver and subdriver scores, an initial screening of the data reveals that data for the Yahoo brand are problematic, because the brand posts a lot of news articles, resulting in 20–40 times more posts than the other internet brands. We thus drop the Yahoo brand from the calculation and subsequent analysis.
Results
We use multiple methods, including descriptive statistics, correlation analysis, and repeated and mixed-measures analysis of variance, to replicate the tracker with the two social media platforms. Table 5 presents the descriptive statistics and correlation analysis for the three social media platforms, along with the overall brand reputation and the three drivers.
Descriptive Statistics and Correlations Across Three Social Media Platforms.
a Denotes the only entry in this table that is not significant at the .05 level.
Notes: Mean, SD, Min, and Max are calculated using raw scores, and correlations are calculated using the normalized scores. Negative value means that the negative sentiment is larger than the positive sentiment.
The descriptive statistics show that Twitter has much higher volume than the other two platforms. This is because Twitter has more publicly available content (500 million tweets daily) and has more brand content than other platforms. Although Facebook also has a high volume of posts (comparable with Twitter), most are not on public pages and thus are not readily available. Instagram’s posts (95 million posts daily) are mostly not brand related. The descriptive statistics confirm empirically our choice of Twitter as the social media platform of the Tracker, as articulated previously.
The correlation analysis shows that the three social media platforms converge at both the brand reputation and the driver levels. All correlations are significantly positive, except the relationship driver between Twitter and Facebook.
A repeated and mixed-measures analysis of variance, by taking into consideration the time-series nature of the measures and the brand differences, further suggests that our tracker can be replicated using other social media platforms. We treat the brand reputation (and its three drivers), respectively, as within-brand repeated measures that are between platforms. We do not find significant platform differences for brand reputation (F = .96, p = .386, d.f. = 2), but we do find significant brand differences (F = 378.63, p = .000, d.f. = 5) and brand × platform differences (F = 87.31, p = .000, d.f. = 10). The results for the three drivers are the same; all platform differences are insignificant, but brand differences and brand × platform differences are significant. The findings confirm that our tracker can be generalized to other social media platforms and suggest that brands have different social media strategies.
Summary
The results of descriptive statistics and correlation analysis provide general support for the robustness of the tracker across social media platforms, which can be text-based or visual-based. Among the three social media platforms, Twitter is more suitable for monitoring and tracking brand reputation, due to its substantially more publicly available brand-related content. Furthermore, the tracker can be replicated using the other two social media platforms at both the brand reputation and the driver levels—evidence for the generalizability of the tracker as a social media–based brand reputation tracker. This replication also supports that our conceptualization and methodology are robust even if the nature of individual social media varies. The results of the repeated and mixed-measures analysis of variance provide stronger evidence by considering the tracker’s time-series nature, social media characteristics, and brand differences.
These multiple approaches to replication consistently support the robustness of the Twitter-based brand reputation tracker: the three-driver framework and the methodology we develop here can be generalized to other social media platforms.
Establishing a Nomological Relationship with YouGov’s BrandIndex
Data
We purchased access to YouGov’s BrandIndex data and matched 71 noninternet brands 7 that are common between the two data sets for the data period (i.e., 130 weeks, 9,230 brand-week observations). YouGov’s BrandIndex interviews a consumer panel about their opinions regarding three broad sets of brand metrics: brand health, media, and purchase funnel metrics. (For a detailed description of YouGov’s methodology, visit https://today.yougov.com/solutions/syndicated/brandindex). Essentially, for the three broad sets of brand metrics, the media metrics are conceptually similar to our tracker. It contains brand WOM (i.e., whether the consumer has recently spoken about the brand) and brand buzz (i.e., whether the consumer has heard anything positive or negative about the brand). The purchase funnel metrics, such as purchase intention, are more appropriate as the nonfinancial outcome of the tracker. Thus, we establish the nomological relationship of the tracker with YouGov’s BrandIndex by conceptualizing its brand WOM and brand buzz as concurrent variables with the tracker, and its purchase intention as the outcome variable, as shown in Figure 7.
Nomological relationship between the tracker and YouGov’s BrandIndex.
Results
A simple correlation analysis shows that the tracker’s overall brand reputation correlates significantly with BrandIndex’s brand WOM (.355, p = .000), brand buzz (.317, p = .000), and purchase intention (.248, p = .000). All three brand reputation drivers of our tracker also all correlate significantly with the three BrandIndex variables.
We then conducted panel regression analysis to establish the causality between the three BrandIndex variables. The results show that brand WOM has a lagged impact (.221, p = .000), while brand buzz has a concurrent impact (.072, p = .027) on purchase intention.
After establishing that the causal chain is likely to be from brand WOM to brand buzz to purchase intention, we ran a panel VAR model to explore the dynamics among BrandIndex’s brand WOM and brand buzz, and the tracker’s overall brand reputation. Results of the Granger causality test show that it is more likely for brand reputation (χ2 = 18.542, p = .000) and WOM (χ2 = 7.690, p = .006) to Granger-cause brand buzz, indicating that the tracker’s brand reputation is likely to be a concept that encompasses BrandIndex’s brand WOM and brand buzz.
Last, we ran a panel regression analysis using the current and lagged-one-week values of the tracker’s overall brand reputation to predict BrandIndex’s purchase intention. Results of the analysis, with standard errors adjusted for brand heterogeneity, show that the lagged brand reputation (not the concurrent one) significantly predicts consumers’ intentions to purchase the brand (.019, p = .017).
Summary
We establish the nomological relationship between our tracker and YouGov’s BrandIndex, with YouGov’s conceptually similar metrics of brand WOM and brand buzz correlating significantly with the tracker, and with the lagged tracker being a significant predictor for consumers’ intentions to purchase the brand.
Converging with Other Aggregate Brand-Related Measures
There are many other brand-related measures, as listed in Table W1 in the Web Appendix. We compare the brand tracker with three other aggregate brand-related measures for 2018: the Interbrand, Forbes, and BrandZ lists (chosen for their availability) to check whether they converge. The correlational analysis shows that the three lists are highly correlated, with correlation coefficients all greater than .851 (p < .000). This indicates that they are very similar, even if they claim to use different methods of brand evaluation. The correlations of the three lists with the overall brand reputation scores are significant (p < .000) but differentiable, because the correlation coefficients range from .079 (Forbes), to .081 (BrandZ), to .127 (Interbrand), indicating our tracker converges with as well as differentiates from the other aggregate measures.
If we look at the ranking, rather than the brand value, the correlation between the three other brand measures is more discriminable. The Interbrand rank is still highly correlated with the Forbes rank (.802) but is more discriminable from BrandZ’s rank (.404). The correlation between Forbes’s and BrandZ’s ranking also become more discriminable (.581). Meanwhile, their correlations with our tracker become higher (.335 for Interbrand; .356 for Forbes; .179 for BrandZ), suggesting that these measures converge better by using rankings.
We further explore the relationship between the 2018 Interbrand ranking and the three drivers. We regress the Interbrand ranking on the three drivers and find that the brand driver predicts Interbrand ranking most closely (.279, p = .000), followed by the relationship driver (.109, p = .015). The value driver does not predict Interbrand ranking, when all three drivers are considered.
In summary, the analysis using the Interbrand ranking provides evidence that the two ranking systems correspond for the brands included, but our tracker provides more granular measures both in cross-sectional dimension (multiple brand drivers and subdrivers) and longitudinal dimensions (weekly fluctuations).The brand reputation tracker not only provides real-time information about a brand’s performance but also is more granular at the driver and subdriver levels, providing additional actionable information about a brand’s performance beyond the typical annual, aggregate-level brand ranking system.
Discussion
We demonstrate the various uses of brand reputation tracker data, using various methods and approaches. Our approach can be used to monitor and manage a brand’s reputation over time, both at the driver and subdriver levels. The data can also be used to manage brand competition by tracking the ups and downs of drivers for major competitors, and the subdriver analyses can provide detailed insights about how to enhance or improve brand reputation. The accountability of the residual of the brand reputation drivers for abnormal returns confirms the financial implications of using the tracker to monitor and manage brands. Finally, the validation against other social media platforms and other brand-related measures provides evidence for the generalizability and external validity of the tracker. Next, we discuss implications for managers and practitioners of the tracker and propose a research agenda for researchers to leverage the availability and the methodology of the tracker.
Implications for Managers
Managing the brand–value reciprocity
A firm can manage the reciprocal relationship between the brand and the value drivers. Depending on which driver a firm has the comparative advantage, the firm can selectively manage one of the drivers first, and then let the effect carry over to the other driver. For example, Apple has a stronger reputation in the brand driver as being innovative, but a weaker reputation in the value driver, due to its premium price; thus, Apple could prioritize leveraging its stronger brand driver (and the innovative subdriver) and let stakeholders understand that the innovativeness is worthy of a premium price (e.g., with better service quality). Figure W4a in the Web Appendix shows that this brand-to-value carryover effect takes one week to take off but lasts for five to six weeks. Alternatively, Samsung has a stronger reputation in the value driver as being affordable while still innovative; thus, Samsung could prioritize leveraging its stronger value driver (i.e., being affordable with high goods quality) and, by doing so, influence stakeholder perceptions of its innovativeness, due to the reciprocal relationship. Figure W4b shows that this value-to-brand carryover effect occurs in real time and lasts for four to five weeks.
Managing the brand–relationship–value virtuous circle
A firm can manage the virtuous circle among the three drivers, from the brand driver to the relationship driver, from the relationship driver to the value driver, and from the value driver back to the brand driver. For example, Apple has a strong foothold on all three drivers (shown in Figures W2a–W2c in the Web Appendix), and thus, it is in a good position to leverage this virtuous circle. For a firm that is good at one or two drivers, it can manage the brand–value reciprocity, as discussed previously. For a firm that is good at the relationship driver (e.g., a mature brand with existing loyal customers but having nothing new to be talked about on social media), the firm can manage the relationship driver to get to the value driver (e.g., generate discussion about its new product, new service, or new price), so that the firm can subsequently leverage the brand–value reciprocity. Figure W4c shows that the relationship-to-value carryover effect takes one week to take off but lasts for five to six weeks, and Figure W4d shows that the brand-to-relationship carryover effect is real-time and lasts for four to five weeks.
Managing brands based on drivers’ temporal impact on financial returns
We find that the three drivers tend to impact a firm’s financial returns at different tempos: The brand driver has a real-time impact, the value driver takes one week, and the relationship driver takes two weeks to play out.
Given that the brand driver reflects brand sentiment that can fluctuate easily with brand events, and given its dominant impact among the three drivers, a firm can boost this driver using brand events and activities, such as a new product launch, and expect a real-time impact on financial returns.
Given that the value driver reflects brand knowledge, which does not change as easily as brand sentiment, a firm can leverage this driver as a relatively more stable foundation of its brand reputation. Its synergy with the brand driver supports this strategy that brand sentiment can be manipulated by brand events, while brand knowledge can settle the sentiment into more enduring brand knowledge that can stabilize the effects of brand sentiment’s ups and downs on financial returns. Brand crisis management is one example: in the case of an unexpected negative brand event (e.g., Facebook’s account data leakage), given stakeholders’ knowledge about Facebook, the impact of the temporarily negative brand sentiment would be settled if a stakeholder has more positive knowledge about Facebook than negative knowledge.
Given that the relationship driver reflects brand relationship, a firm can leverage this driver for long-term returns that are less subject to temporal fluctuations, though the firm needs to be patient with branding efforts for building relationship. Although the observation that relationship takes time to play out is established in the existing customer relationship literature, its long-term synergy with the other two drivers is not yet widely recognized and can be leveraged for a stable brand reputation and its impact on financial returns.
Implications for Data Trackers and Providers
Our theory-data iterative approach to developing the tracker illustrates the importance of theory and the need for data providers to pursue academic collaborations. Most data providers that track data or provide raw or summary data do not have a sound theory to guide them about what data to track and what summary data would be valuable. For example, Crimson Hexagon tracks data but does not provide raw data, whereas YouGov provides data but allows clients to provide guidance on what data to collect. With the black-box machine learning approach continuing to advance for data tracking, making sense of data will be a pressing issue. Our methodology illustrates the importance of theory-based data tracking.
Implications for Researchers
The longitudinal brand data are available for free access to the academic research community. The data can be used in a variety of ways. In this section, we provide a list of possible research agendas along with a sampling of specific research questions for future research that can successfully leverage the data. Table 6 lists the research agendas and specific research questions. We discuss these in detail next.
Agenda of Future Research Questions Using the Tracker.
Social media lens to brand reputation
The tracker is built using social data from Twitter and is demonstrated to be generalizable to other social media platforms, which uniquely reflects how stakeholders talk and think about brands on social media. Given the growing importance of social media mentions for brands, one potential use of the data is to understand brand reputation on the basis of social media activities. In the example analyses, we find that stakeholders talk about Samsung as “cool,” whereas Pepsi is not “cool.” Such wordings are distinct from how brands are portrayed in traditional media but are caught uniquely by the tracker.
Cross-platform application
The tracker is applicable to multiple social media platforms that are distinct in data type (e.g., text, photo) and interaction pattern (e.g., one to many or one to one). Although different platforms have very different purposes (e.g., Facebook is mostly for social interaction between friends, Twitter is more of a “broadcast” platform), we find that our approach produces meaningful results even for very different platforms.
Longitudinal research on brand reputation
Given the time-series nature of the data, one area of research that can make use of the tracker is to explore variation in brand drivers and subdrivers over time. This is important because drivers of brand reputation are unlikely to be stable over time for all brands and in fact might vary considerably. The within-brand (over time) and between-brand temporal volatilities and dynamics of these measures are worth exploring as a use of the data. Ultimately, the dynamic richness of the data should open up new empirical possibilities for researchers interested in understanding how brands evolve, in a multitude of ways, between brands, across categories, and over time.
Granular investigation of brand reputation
The granularity of the tracker in both the time (weekly, monthly, and quarterly) and driver (drivers, subdrivers, and sentiments) dimensions enables researchers to examine brand reputation in a much more granular way. On this front, the numerous drivers and subdrivers allow for novel theory building and testing opportunities in relation to impacts of exogenous shocks on brands. For example, a theory could predict an impact of a shock or event on one driver but not another, and empirical evidence for this could be sought using our tracker, where there are theoretically meaningful differences between the affected and unaffected drivers that (indirectly at least) shed light on the underlying mechanism for an observed effect.
Brand-related events for brand reputation variation
Another potentially fruitful avenue for researchers is to identify brand-related exogenous events for brand reputation variation over time to see how specific events that are relevant to given brands in the tracker affect brand reputation (overall and for each of the drivers and subdrivers). One obvious application is to examine how brands are impacted by negative and positive exogenous events such as product recalls, crises and scandals, major announcements, changes in C-level executives, product launches, and other potentially significant strategic marketing actions. Although prior research has at times considered such topics, including in the context of social media (e.g., Borah and Tellis 2016), more work is needed to increase our understanding, in the finer granularity provided by the subdrivers.
Brand, stakeholder, and firm characteristics for brand reputation driver variation
One theoretically fruitful area is to link subtle and interesting brand, stakeholder, and firm characteristics to the variation in the brand reputation drivers. Using those data to account for the fluctuations of brand reputation drivers allows researchers to develop new models and theories about why a certain brand “attracts” or “offends” stakeholders on social media, as a function of those characteristics. We demonstrate a nomological relationship of the tracker with YouGov data, which can be one of the many approaches to link the driver variations to those characteristics.
Brands in novel classifications
One potential way that researchers can benefit from our tracker is to classify brands in line with their scores on the three brand reputation drivers or the larger set of subdrivers, or on the basis of statistically estimated properties such as the extent to which time series within brands are correlated/cointegrated, stationary/evolving, and so on. In addition, machine learning techniques for classification could be used to achieve a similar outcome. This can lead to the identification of new types of classes or groups for brands that might have interesting theoretical and practical implications. This could also lead to new research questions pertaining to understanding differences between brands that score high versus low on various drivers of interest.
Brand reputation drivers and marketing/financial outcomes
We demonstrate some applications of the data by linking brand reputation drivers to firm abnormal stock returns. Our analysis illustrates the potential available for researchers interested in the marketing–finance interface. Given the time-series nature of our data, more complex models could be developed that allow for marketing/financial metrics not only to be considered as being influenced by brand reputation but also to have an effect on changes in the various drivers in our data. This would be interesting, as it would enable us to understand the extent to which brand reputation drives, for example, financial outcomes versus how much reputation is instead driven by firm performance.
Contribution
As opposed to most aggregate ratings or rankings of brand value, the brand reputation tracker enables a more granular investigation of the components of brand reputation. Unlike survey-based attitudinal brand measures, the tracker is designed to map more directly to competing strategic marketing expenditures.
We tie to the expanding literature on mining (“listening in on”) social media to obtain brand reputation insights. By developing a methodology that is applicable across multiple social media platforms and providing a longitudinal database that is granular enough to guide marketing actions, we make it much easier for researchers to tie social media posts to brand reputation.
We provide a new resource for corporate reputation research. To date, most corporate reputation research has been in the management and strategy literature streams and has placed less emphasis on marketing actions. This database enables corporate reputation research to link more naturally to marketing.
Finally, we contribute to the research of brand reputation by making our brand reputation tracker data available to facilitate longitudinal brand research. Further extending over additional time periods or additional brands could be a valuable resource for future brand research and increase our understanding of how brands work.
Limitations
Our tracker is based on Twitter tweets. This gives the tracker many advantages over other social media. However, the social media environment is not static; for example, Twitter might go out of business, or a change in its terms of service might make it impossible to apply our brand tracker on the platform. In such an eventuality, researchers may need to migrate the tracker to a different social media platform. We have demonstrated the generalizability of the methodology using data from Facebook and Instagram, suggesting that our approach may be usable on other platforms as well.
Historical data are more difficult to collect. We suggest that brands that are not included in the tracker or that want to add more actionable subdrivers should be forward-looking by following our methodology and accessing the underlying data using the public free streaming application programming interface to build their own tracker. Purchasing historical data from data providers is still an option, as we did to backfill the missing data.
Although we can mine millions of tweets automatically, there is still the need to update the usernames manually. Our collection is limited to English tweets. People’s use of keywords for talking about brands may also change over time and across contexts, and thus the dictionaries need to be updated periodically. We expect that with more advanced machine learning, usernames and subdriver dictionaries can be updated automatically.
We start from 14 subdrivers and refine them into a smaller set of 11 subdrivers based on the multiple-stage, theory-data iterative process. These subdrivers are shown to be applicable to brands in the data set. Subdrivers are directly actionable, and thus, a firm can explore more potentially actionable subdrivers for further enhancing the marketing relevance of the tracker to its brand.
Conclusions
The brand reputation tracker is a longitudinal data base of brand reputation driver and subdriver data, for 100 top global brands, based on mining Twitter tweets. Our study contributes to the literature by making brand reputation financially accountable and managerially actionable in real time and over time. The tracker is highly time-sensitive and context-specific, allowing firms to respond quickly to market stimuli. The final goal is to provide a database resource that any academic researcher can access and/or extend. We anticipate that this should increase the amount of research done on brand reputation over time, increasing our knowledge of the antecedents and consequences of the components of brand reputation. We also hope and expect that the tracker will give marketing more importance in the broader corporate reputation literature.
Supplemental Material
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921995173 - Real-Time Brand Reputation Tracking Using Social Media
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921995173 for Real-Time Brand Reputation Tracking Using Social Media by Roland T. Rust, William Rand, Ming-Hui Huang, Andrew T. Stephen, Gillian Brooks and Timur Chabuk in Journal of Marketing
Footnotes
Acknowledgments
Thanks to OssaLabs for its collaboration in building and maintaining the database, to YouGov for access to data used in this study, and to James Davies and the IT staff at the Oxford’s Säid Business School for their help in setting up the website. The authors would also like to thank Bradley Senft, who helped with the development and implementation of the software to collect and process the social media. They also thank Rupert Younger, Rowena Olegario, and participants at the Oxford Corporate Reputation Forum and the American Marketing Association Winter Conference for their helpful comments
Author Contributions
Roland T. Rust, William Rand, and Ming-Hui Huang contributed equally to the article.
Associate Editor
Tomas Hult
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the University of Oxford’s Centre for Corporate Reputation, and the Center for Excellence in Service at the University of Maryland, grants (104-2410-H-002-142-MY3, 106-2410-H-002-056-MY3, 107-2410-H-002-115-MY3) from the Ministry of Science and Technology, Taiwan, and financial support from E.SUN Commercial Bank.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
