• No results found

4. Data and methods

4.2 Operationalization of variables

The current study gathered field data about firms and users’ activities and interactions on firms’ fan sites [Twitter and Instagram accounts] to capture the mutual communication between the firms and their [potential] customers. The variables are divided into two categories, namely: firm-centric and user-centric. In the study’s conceptual framework (Figure 3) the independent and moderating variables are the firm-centric variables. Those capture companies’ efforts in content generation (Shahbaznezhad et al., 2021). The

37 independent variables allowed us to investigate how the companies manage their digital content. On the other hand, the dependent variables show users’ reactions to firms’ content.

These are the user-centric variables and were measured in the form of customer engagement on social media fan pages (Hoffman & Fodor, 2010). The customer engagement in this research was measured by likes, comments and sentiment of the comments. The operationalization of the variables is shortly given in Table 8.

Independent variables

As mentioned before, the independent variables that were used in the current research are the type of content, divided into three sub-categories (informative, persuasive, and entertaining), platform, product type, and product category as well divided into three sub-categories

(fashion, electronics, and travel). A content analysis has been performed to operationalize the independent variable Type of Content. Content analysis is defined as

“analysis of the manifest and latent content of a body of communicated material (as a book or film) through classification, tabulation, and evaluation of its key symbols and themes in order to ascertain its meaning and probable effect.” (Krippendorff, 2018).

A qualitative approach was employed for the content analysis, which means that the social media postings of the monitored firms were carefully reviewed and assigned to different sub-categories. The posts were given codes by deconstructing the text into three categories:

informative, persuasive, and/or entertaining. Subsequently, the measures chosen for each variable fitted the unit of analysis’ conceptualizations as previously established in the literature review. All 240 posts were being intensively read and coded during the study. The content was coded in a binary manner meaning that the presence of a given sub-category was classified as 1, while its absence was coded as 0. As a result, an extensive overview was

38 created, with each posts’ measurements and codes shown alongside it. Posts may be rated 1 or 0 for the independent variables Type of Platform and Type of Category however, there were no chances of rating 1 for both sub-categories. Because nearly no posts had textual content for a product or brand that was both hedonic and utilitarian, the posts for the independent variable Type of Product were also coded either 1 or 0. On the contrary, it is possible that two or even three types of content could be identified in one post, so the maximum number of content items that any post had was 3; in other words, a post was coded with 1 for each content sub-category.

Dependent variables

The current study relied on a Web Scraping approach to operationalize the dependent

variables. The number of likes and comments was calculated by counting the platform users’

activities for each post. Table 7 provides a summary statistic of the outcome variables.

In total, this research used four different dependent variables: Ratio Likes/Followers, Ratio Comments/Followers, Engagement Rate (%), and Sentiment Score. Using multiple models with different outcome variables in a comparative manner, the current research considered different engagement outcomes into that might be triggered-influenced by different predictors or moderators. In total there were four models, each with a different engagement outcome, while these models had a similar predictor – moderator variables (X) set. Thus, only the outcome variable (Y) had been changed while running each model in the analysis stage. In the discussion section of the report, these differences are focussed on further.

39 Table 7

Summary statistic of dependent variables

.

Ratios were calculated since the observed firms and their linked social media profiles had a varying number of followers. The number of comments was divided by the number of followers to create a dependent variable called ratio comments/followers. Similarly, the number of likes was altered. In addition, the engagement rate was calculated by adding the total number of comments and likes on a post dividing this number by the number of

followers. The percentage was then calculated by multiplying this by 100. The ratios came out as too small values, which is very natural, however for this to not cause any problems when put-used in regressions and while running the analyses, the variables were twisted by multiplying them with 10,000.

To calculate a sentiment score for each brand post, the automatic process sentiment analysis of MonkeyLearn has been used. Each comment on a firms’ post has been run through MonkeyLearn to get a positive (1.0), neutral (0.5), or negative score (0.0). Afterwards, the scores for all comments of the brand post were added and divided by the total number of comments resulting in a sentiment score between 0 and 1 where 0 represents the most

Number of likes Number of comments Sentiment Score of comments

Twitter Louis Vuitton Sum 535,927 4,447 -

Average 13,398 111 0.36

Dell Sum 1,348 327 -

Average 34 8 0.13

KLM Sum 15,986 821 -

Average 399 21 0.28

Instagram Louis Vuitton Sum 4,300,249 27,026 -

Average 107,506 676 0.52

Dell Sum 92,848 3,090 -

Average 2321 77 0.53

KLM Sum 431,461 9,629 -

Average 10,787 241 0.64

40 negative score compared to 1 for the most positive one. As the usage of Emoji in text mining and sentiment analysis could affect the sentiment score (Ayvaz & Shiha, 2017), those were considered within the analysis.

Moderating variables

Moderating variables influence the strength or nature of the relationship between two

variables. Generally, a moderator is any variable that affects the relationship between two or more other variables, whereas moderation is the moderator’s effect on this relationship (Dawson, 2013). The current research hypothesises that the type of platform, type of product, and product category are potential moderating variables that affect the direct relationship between the variables type of content and customer engagement. Due to a large number of sub-categories of independent variables in the framework, dummy coding was employed for all three moderating variables to reduce the effect of unwanted variance.

Type of platform addresses the effect of the platform that has been used to post the content online. Previous research has found that the type of platform performs an important role in facilitating social media engagement behaviour (Shahbaznezhad et al., 2018). For the type of platform, posts placed on Instagram were labelled 1 and posts placed on Twitter as 0.

The second moderator was the type of product. Multiple studies found significant evidence that hedonic and utilitarian values are essential drivers of customer engagement

(Zyminkowska, 2018; Kim et al., 2019; Huang & Rust, 2021). The current study assumes that informative or persuasive content for utilitarian products would generate higher customer engagement outcomes. Subsequently, it is hypothesized that entertaining content improves customer engagement outcomes for hedonic products. A dummy variable has been created in which hedonic products were labelled 1 and utilitarian products as 0.

41 Table 8

Operationalization of the variables

Operationalization Scale

Independent Variables

Informative content Sub-variable where presence of the type of content is labelled 1 and absence as 0

Binary, 0 or 1

Persuasive content Sub-variable where presence of the type of content is labelled 1 and absence as 0

Binary, 0 or 1

Entertaining content Sub-variable where presence of the type of content is labelled 1 and absence as 0

Binary, 0 or 1

Dependent Variables

Ratio likes/followers The ratio is calculated by dividing the quantity of likes on the post by the total amount of followers

Numerical

Ratio comments/followers The ratio is calculated by dividing the quantity of comments on the post by the total amount of followers

Numerical

Engagement rate (%) Calculated by adding the total number of likes and comments on the post, divide this by the total amount of followers, and multiply the number by 100

Numerical

Sentiment score Rated each comment as 1 for positive, 0.5 for neutral, and 0 for negative by using Monkey Learn. Subsequently, the sentiment score of a post was calculated by adding the scores and dividing them by the total number of comments of the post.

Numerical, precise number between 0 (most negative) and 1 (most positive)

Potential Moderators

Type of platform Dummy variable – Instagram (1) or Twitter post (0)

Binary, 0 or 1

Product type Dummy variable – Hedonic (1) or Utilitarian product (0)

Binary, 0 or 1

Type of category Two dummy variables where electronics category is the basis category

- Category dummy Fashion (1) where electronics and all others are the base (0) - Category dummy Travel (1) where

electronics and all others are the base (0)

Binary, 0 or 1

42 The Product Category is the third and final moderating variable evaluated in this study. The category type is determined by selecting firms from three categories: fashion, electronics, and travel. This empirical research used a dataset generated from the three significant companies operating in those three categories to test the conceptual framework. Two dummies were created for this variable, for which the electronics category was used as the base category.