BUS631: Brand Management

Research Background

In recent years, many scholars have generally acknowledged the key role of social media in brand development and emphasized the positive impact of brand engagement behaviors on digital customer engagement. We collected a dataset from the social media platform Weibo. First of all, Weibo is the most popular open social media platform in China, with 511 million monthly active users according to the Weibo 2020 User Development Report, maintaining an obvious advantage over other platforms. Secondly, Weibo users can disclose personal or brand information such as interests, complaints, brand preferences, and brand-related experiences through text, images, and videos. Lastly, most of the social network studies used in research are based on or include Weibo data. Thus, Weibo becomes the most relevant social network for consumers and brands.

In order to emphasize brand generalization and comparison, this paper selects Brand Finance's list of Top 500 Brands in China 2021 as a brand selection criteria to further take in effective social media strategies. This sampling framework has several advantages over lists on other business lists such as Fortune. First, since other lists cover brands on a global scale, cross-cultural factors such as national sentiment or cultural differences might lead to additional influence, therefore, choosing this list is more appropriate for this study to investigate from the same cultural context. Second, the top-ranked companies on the list are involved with multiple industries, including technology, media culture, airlines, and restaurant business, which may bias the findings toward two discretionary purchases types. Third, most of the brands on the list post relevant information on Weibo, which makes Weibo a practically important approach to get to the consumers.

Sampling and Data Collection

Brand Selection

Based on Brand Finance's assessment, this study selected the brands by following the criteria below. First, to prevent industry homogeneity, multiple categories of industry representatives were selected. Second, it has been observed that many brands have set up more parallel accounts based on sub-brand or product categories. In order to maintain consistency and comparability with other brands with unique accounts, we chose to collect data about the company's brand level. Lastly, brands with less than 6 months of Weibo postings and low activity are filtered out.

Before the formal survey, discretionary purchases were investigated for brand screening using an online questionnaire, and a total of 89 online questionnaires were collected. In the pretest, information on several real companies from 18 different industries (e.g., media culture, apparel, cosmetics, banking, airlines, technology, communications, etc.) were provided. Building on the research of Van Boven and Gilovich and van Doorn et al., respondents were asked to think about their purchases experience and to choose a type of discretionary purchases within the industry to maximize this difference (e.g., a plane ticket vs. a refrigerator). They determine the type of purchases on industry attributes by choosing options like "experiential purchases," "material purchases," "unsure" or "refused to answer," to indicate their perceptions of both experiential and material purchases categories. Based on the survey results, a total of four Chinese brands with predominantly experiential purchases were selected: Air China, China Southern Airlines, Tiktok, and Tencent; and the four Chinese brands with predominantly material purchases: SAIC Motor, Geely Auto, Xiaomi, and Huawei. These eight companies have high brand familiarity and a relatively comparable number of brand followers in 2021, and post a similar amount of content on Weibo during 2020 and 2021.

Data Collection

The time span for data collection in this study was from June 1, 2021 to August 31, 2021. Firstly, this time span increases the longitudinal nature of the dataset, which helps to further reduce the prevalence problem. Secondly, it allows us to mitigate additional effects due to transient activity events. We use python to collect numerical customer engagement related metrics for all posts and their corresponding likes, shares, and comments during the time span. After data cleaning, the final dataset contains 1,519 social media posts, 40,142 comments, etc. Due to the default settings of Weibo and the limitations of the data crawler, we only collected about 70–80% of the top-ranked comments.

Operationalization of Variables

In order to capture the digital engagement between the company and its current users, we collected data about the company and user activities and interactions on the official brand Weibo page. Our variables were divided into two categories: (i) brand-centric and (ii) user-centric. The brand-centric variables are independent and moderating variables in the study's conceptual model to capture the brand's ongoing efforts on social media. On the other hand, user-centric variables (dependent variables) show the extent to how users respond based on the brand's social media strategy, i.e., digital customer engagement.

Independent Variables

Brand social media content strategies

Based on the literature and the theoretical framework presented in the previous section, we operationalized and tested three content strategy categories by measuring brand social media content strategies as categorical variables, namely 1 = "Information strategy"; 2 = "Community strategy"; and 3 = "Action strategy." Two trained coders coded each blog post of the brands. If a Weibo post of a certain strategy type existed, it was coded as 1, while posts with no such content was marked as 0.

Brand social media response strategies

This study is carried out on the research basis of Kent and Taylor and Rourke et al. to explore the extent to which brand social media response strategies reflect the brand social presence. It is also based on three communicative strategies and related metrics developed by Rourke et al. that contribute to social presence. Specifically, affective response can be expressed on social media through (i) emoji; (ii) humor; and (iii) voluntary disclosure.

Interactive response refers to explicitly identifying other communication partners' messages by (i) continuing the conversation by explicitly referencing other messages, (ii) asking questions, (iii) expressing appreciation/compliment, and (iv) expressing consent.

Cohesive response uses (i) name to address people or reference to members of the public (both internal and external); (ii) inclusive pronouns to address or refer to groups (e.g., we, community, society); and (iii) purely social features, greetings, and other social techniques to maintain brand sentiment.

Brand image

Brand image is a dummy variable divided into "warmth" and "competence" (0 = competence, 1 = warmth).

Discretionary purchases

To determine whether there is a purchases type difference in the impact of brand social media response strategy on digital customer engagement, discretionary purchases is set as a dummy variable divided into experiential purchases and material purchases (0 = material purchases, 1 = experiential purchases).

Dependent Variables
Positive filtering, cognitive and affective processing, and advocacy represent the different levels of digital customer engagement. The number of likes, comments and shares is calculated by counting user activity per post.

Positive filtering

"Likes" is the most common engagement practice on Weibo. According to the above definition, "like" is a form of positive evaluation of branded content. Therefore, we describe hitting the "like" button as a weak form of DCE but showing a positive emotional state toward the brand or the content.

Cognitive and affective processing

We measured the number of consumer comments as a primary metric of cognitive and affective processing. Consumer comment is a good metric of the level of digital customer engagement. On Weibo, the number of comments represents the opportunity for consumers to see brand posts, reflecting their level of cognitive and affective processing.

Advocacy

"Shares" plays an active role in spreading branded content and serves as a reliable source and the most powerful form of DCE. Users tend to identify themselves, develop social relationships and influence others by providing information about the brand in their social networks and sharing content with others and different social groups. Therefore, we counted the number of shares to measure the advocacy dimension.

Control Variables
Next we examine the control variables in the model. We include these variables below to control for heterogeneity between brands, which could explain some of the observed differences in the impact of brand and digital customer engagement associations.

Time control variables

First, Weekdays. Weekend post has a significant impact on enterprise content strategy, and we use it to capture weekend and weekday seasonality. Weekdays is a metric variable to determine whether posts occur on weekends (=1) or weekdays (=0). Second, the posting schedule. Posting schedule is considered as a key consideration for account managers when designing social media strategies, and the timing of posting predicts customer engagement. Therefore, our posting schedule variables were dummy coded to divide the time into two parts: posts from 8 a.m. to noon and 2 p.m. to 5:30 p.m. (working hours) were marked as 1, and posts from noon to 2 p.m. and 5:30 p.m. to 8 a.m. (resting time) were marked as 0.

Brand control variables

The level of digital customer engagement may vary with the number of brand followers vs. the level of brand activity. Given the study by Dhaoui and Webster, this paper adds three variables, which are the number of followers, overall posting volume and posting frequency, to observe the extent of their differences on digital customer engagement.

Media richness

Media formats include the corresponding ability to deliver messages containing rich information. This paper takes into consideration the objective characteristics of media formats that not only determine their ability to spread information but also trigger different levels of digital customer engagement. Building on the research by Tseng et al. and Shahbaznezhad et al. and others, media richness was operationalized as a categorical variable consisting of four dimensions:(i) topics; (ii) content and links; (iii) photos and images; and (iv) videos.

Length of content

Previous research has found that message length may positively or negatively affect outcome metrics. Therefore, we controlled the length of the content of the blog posts.

Coding Procedures
To avoid subjective bias, the eight selected companies were randomly divided into two groups to be handled by two coders. Both coders received detailed training on the coding tool and classification set, and the coding process took approximately 2 weeks. The independent variable metrics were all coded according to a dichotomy of information presence (i.e., 1 or 0) to minimize possible subjective decisions by the coders. Inter-coder reliability was then calculated. We measured Cohen's Kappa value. The value of Cohen's Kappa for brand social media content strategy is 0.797, the value of Cohen's Kappa for brand social media response strategy is 0.846, and the value of Kappa for brand image is 0.855, suggesting "substantial agreement" between the coders. In addition, some of the Weibo content in the extended dataset involving new message types were re-evaluated, requiring only the adjustment of existing subcategories rather than the construction of new subcategories.

Basic Model
In order to test the hypotheses, we used a multiple linear regression model with categorical variables. To explain positive skewness, our first step is to add 1 for log-non-linear transformation by calculating the natural logarithm of likes, comments, and shares to avoid the possibility of taking log 0 as the dependent variable in the analysis (Ba and Pavlou, 2002). That is,

$Y_1=lnlikes; Y_2=lncomments; Y_3=lnshares$

Y_ij [y_i1 or y_i2 or y_i3] is the number of likes or comments or shares for each brand post i. Controli is a control variable, including weekdays, posting schedule, media richness, etc.

For each blog post i, we formulated the basic regression model as follows:

$\begin{aligned}& \mathbf{Y}_{\mathbf{i} j}=\alpha_{i j}+\beta_1 \text { content strategy } y_{i j}+\beta_2 \text { Affec_response }{ }_{i j} \\& +\beta_3 \text { Inter_response }{ }_{i j}+\beta_4 \text { Co_response }{ }_{i j} \\& +\beta_5 \text { Discretionary purchases }_{i j}+\beta_6 \text { Brand }{\text {image }}{ }_{i j} \\& +\beta_7 \text { Control }_i+\varepsilon_{i j} \\&\end{aligned}$

To see whether discretionary purchases and brand image have a moderating effect on the relationship between brand social media response strategies and the overall level of digital customer engagement, based on previous research, we used raw scores of likes, comments and shares, calculated additional dependent variable weights and took the natural logarithm as the overall level of digital customer engagement.

$\mathrm{l} n D C E=0.5^* \sum(\mathrm{L})+0.5^* \sum(\mathrm{C})+0.5^* \sum(\mathrm{S})$

$\mathbf{l n D C E}_{\mathbf{i}}=\alpha_i+\beta_1 Response \; strategy _i+\beta_2 Discretionary \; purchases _i +\beta_3 Brand {\text { image }} i+\beta_4 Response \; strategy _i{ }^* \\ Discretionary \; purchases { }_i+\beta_5 Response \; strategy { }_i{ }^* \\ Brand \; {\text {image }_i}+\beta_6 Control _i+\varepsilon_i$