Different tests and analyses have been run to explore the proposed research hypotheses within the three product categories and for the two product types and platforms.
Multicollinearity Checks
Surveys and experiments were not part of the current research’s data collection, which means there aren’t any Likert scale measurements involved. However, multicollinearity could occur.
Multicollinearity occurs when two or more predictor variables are correlated with each other.
As a result, the standard error of the coefficients increases. Subsequently, increased standard errors indicate that the coefficient for some outcome variables may be significantly different from zero. In other words, multicollinearity overinflates standard errors, making some of the variables statistically insignificant when they should be significant (Daoud, 2017).
Multicollinearity has been checked for all four predictor variables on the four outcome variables.
The output of the multicollinearity detection is given in Appendix 1 – Multicollinearity DetectionFor all of the predictor variables, Tolerance Statistics > 0.2 and Variance Inflation Factor < 5, which means there are no collinearity problems. In other words, there is no correlation between any of the individual predictor variables detected. Thus, the independent variables are not strongly related.
Regression Analysis
To verify the hypotheses of the current study, a regression analysis has been done. Regression analysis is one of the most extensively utilized strategies for assessing multi-factor data. The conceptually reasonable approach of employing an equation to explain the relationship between a variable of interest, the outcome or dependent variable, and a set of related
44 predictor variables accounts for its broad appeal and use (Montgomery & Peck, 2021).
Regression measures which of the independent variables actually have an effect on the dependent variable(s). Regression analysis aims to create mathematical models that describe or explain possible correlations between variables (Seber & Lee, 2012). In other words: the research examined how much of the variance in the data can be explained by the independent variables.
The multiple regression model estimates beta coefficients while holding all other variables constant. This means that it assumes that the effect for an independent variable is the same at all levels of the other independent variables (Nayak, 2020). A regression model does not indicate that the independent and the dependent variable(s) have a cause-and-effect
relationship. Even if a strong correlation between two or more variables has been found, this isn’t necessarily evidence that the regression variables and the dependent variable(s) are related in a cause-and-effect manner (Montgomery & Peck, 2021).
Main Effects Analysis
At first, the main effects of each individual predictor variable were tested through a multiple regression analysis. The outcomes resulted in the main effects of each predictor variable. As mentioned before, a regression model was run for each separate outcome variable, so four models were run in total. All four models made sense as a whole with a significance level of p
< 0.001. Another interesting outcome of running those four regression models is that the predictor variables have the most negligible effect on ratio comments/followers, as this model has the lowest R2-value. In other words, the slightest variance in the outcome ratio
comments/followers could be explained by the predictor variables of this study.
45 Table 9
Main Effects regression
R Square F Sig.
Model 1 Ratio likes/followers as DV .271 12.327 .000
Model 2 Ratio comments/followers as DV .144 5.581 .000
Model 3 Engagement rate as DV .275 12.578 .000
Model 4 Sentiment score as DV .325 15.973 .000
Note. N = 240
The relationship between Type of Platform and the four different outcome variables
[sentiment score, ratio likes/followers, ratio comments/followers, and engagement rate (%)]
was studied. For all dependent variables, the unstandardized beta of the predictor variable Type of Platform is statistically significant with a significance level of p < 0.01, suggesting that there is a significant difference between the online platforms Twitter and Instagram such that content placed on the platform Instagram generates higher customer engagement values (i.e. higher ratio likes/followers, ratio comments/followers, sentiment score, and customer engagement (%)) compared to content placed on Twitter.
Subsequently, for some dependent variables, the unstandardized beta value of the predictor variable persuasive content is statistically significant on a p < 0.05 significance level. This is the case for both outcome variables ratio likes/followers and engagement rate (%). The
unstandardized beta values are β = -16.164, p < 0.05 and β = -1655.363, p < 0.05 respectively, meaning that persuasive content has a statistically significant negative effect on the outcome variables ratio likes/followers and engagement rate.
Also, the predictor variable product category had some significant effects on the engagement outcome variables. When content is posted online for firms in the travel category, the content generates higher outcomes on ratio likes/followers, engagement rate (%), and sentiment score compared to content posted online for firms in the electronics category. Those correlations
46 were significant on a p < 0.01 significance level. Besides, content posted online for firms in the fashion category significantly results in a lower ratio of comments/followers with a significance level of < 0.05 and an unstandardized β-value of -0.645 than digital content posted for firms in the electronics category. However, it creates significant higher outcome (0.364) on the dependent variable sentiment score with a significance level of p < 0.001, meaning that digital content posted for firms in the fashion category creates a more positive sentiment score than digital content posted for firms in the electronics category.
Moderation effects analysis
Multiple regression has been used to examine the moderation effects on the dependent
variables. More specifically, multiple regression models were rerun with categorical variables.
Table 10 presents the results of the regression analyses that were run for all four dependent variables. A distinction between two models has been made in the table for each outcome variable. The first model represents the regression model’s results with the main effects of the independent variables. The second model shows the interaction effects of the main
independent variables. Interaction effects are defined as the difference between groups on one treatment variable depending on the level of the second treatment variable (Shahbaznezhad, 2018).
Again, all four models made sense as a whole with a significance level of p < 0.001. Also, the predictor variables and the interaction variables had the most negligible effect on ratio
comments/followers, as this model has the lowest R2-value. For each model, the R2-value slightly increased, meaning that the interaction models explained a little more of the variance in the dependent variables.
47 Table 10
Summary statistic of main effects and interaction effects regression
Sentiment Score Ratio Likes /
Followers Without moderation
effects model
With moderation effects model
Without moderation effects model
With moderation effects model
R square 0.325 0.338 0.271 0.298
F 15.973*** 8.204*** 12.327*** 6.816***
(Constant) 0.147* 0.138 -11.726 4.095
Independent variables
Type of content Informative -0.014 -0.029 3.820 11.981
Type of content Persuasive -0.058 -0.027 -6.016 -8.906
Type of content Entertaining 0.061 0.052 0.888 -15.660
Platform (Instagram = 1 vs.
Twitter = 0)
0.195*** 0.197** 39.746*** 38.605***
ProductType (Hedonic = 1 vs.
Utilitarian = 0)
-0.020 -0.024 9.824 -19.166
ProductCategory (Fashion = 1 vs.
Electronics and all others =0)
0.364*** 0.371*** 1.806 9.125
ProductCategory (Travel = 1 vs.
Electronics and all others = 0)
0.137** 0.109* 24.387*** 30.012***
Interactions (moderations)
Instagram x Entertaining - -0.001 - 0.770
Hedonic x Entertaining - 0.048 - 24.919
Hedonic x Entertaining x Fashion - -0.049 - -1.989
Utilitarian x Informative 0.038 -20.808
Utilitarian x Informative x Fashion (Electronics as basis = 0)
- - - -
Utilitarian x Informative x Travel (Electronics as basis = 0)
- 0.093 - -14.756
Utilitarian x Persuasive - -0.070 - 12.880
Utilitarian x Persuasive x Fashion (Electronics as basis = 0)
- - - -
Utilitarian x Persuasive x Travel (Electronics as basis = 0)
- -0.234 - -14.170
48 Ratio Comments /
Followers
Engagement Rate (%)
Without moderation effects model
With moderation effects model
Without moderation effects model
With moderation effects model
R square 0.144 0.179 0.275 0.302
F 5.581*** 3.510*** 12.578*** 6.940***
(Constant) -0.009 -0.249 -1173.513 384.551
Independent variables
Type of content Informative 0.127 0.188 394.667 1216.851
Type of content Persuasive 0.018 -0.107 -599.782 -901.334
Type of content Entertaining -0.409 0.072 47.829 -1558.773
Platform (Instagram = 1 vs.
Twitter = 0)
0.922*** 1.269*** 4066.763*** 3987.462***
ProductType (Hedonic = 1 vs.
Utilitarian = 0)
0.455 0.965 1027.873 -1820.089
ProductCategory (Fashion = 1 vs.
Electronics and all others = 0)
-0.645* -1.181* 116.138 794.391
ProductCategory (Travel = 1 vs.
Electronics and all others = 0)
0.413 0.670* 2480.039*** 3068.182***
Interactions (moderations)
Instagram x Entertaining - -0.585 - 18.503
Hedonic x Entertaining - -0.939 - 2397.927
Hedonic x Entertaining x Fashion - 1.040* - -94.954
Utilitarian x Informative - -0.002 - -2080.975
Utilitarian x Informative x Fashion (Electronics as basis = 0)
- - - -
Utilitarian x Informative x Travel (Electronics as basis = 0)
- -0.767 - -1552.313
Utilitarian x Persuasive - 0.328 - 1320.791
Utilitarian x Persuasive x Fashion (Electronics as basis = 0)
- - - -
Utilitarian x Persuasive x Travel (Electronic as basis = 0)
- -0.006 - -1417.571
Note. N = 240. Significance level of significant variables: ***p < 0.001, ** p < 0.01, *p < 0.05.
The statistical tests show that the interaction between type of product, type of content, and type of category, in other words: hedonic x entertaining x fashion, is significant on a p < 0.05
49 significance level. More specifically, entertaining content posted for hedonic products and within the fashion category significantly generates higher outcomes (1.040) on the dependent variable ratio comments/followers as shown in Table 10.
Results
Following the analyses presented in the previous section, the integrity of each proposed research hypothesis is assessed in this paragraph. Table 11 gives a quick summary of which hypotheses are supported and which are not.
The output of the performed statistical tests, which is given in Table 10, suggests that entertaining content generates a more positive sentiment score which would partly support H1. However, no significant effects were found for the direct impact of the sub-categories type of content on the four different outcome variables. As proposed in H1, entertaining content was expected to have a more substantial and more positive impact on customer engagement than informative/persuasive content. The second prognosis in H1 was that this effect would be even stronger on Instagram in comparison to Twitter. In the output, one can see that the interaction variable Instagram x Entertaining does generate stronger and positive outcomes for ratio likes/followers and engagement rate (%) compared to the entertaining content variable on its own; however, the opposite is true for ratio comments/followers.
Thereby, none of those effects are significant meaning that entertaining content and the platform on which the content is presented do not have any significant interactions. In other words, entertaining content and its effect on customer engagement outcomes are platform-independent. Thus, H1 of the current research could not be supported.
The findings of the analyses, on the other hand, show that the social media platform
environment of a company’s fan page has a substantial impact on customer engagement. As previously mentioned, the platforms Instagram and Twitter were used for this research to
50 study the moderating role of the variables. The results confirm that firms’ social media profile followers tend to generate higher and more positive outcomes on customer engagement for content placed through the platform Instagram. In addition to that, fan page users tend to
‘like’ more than to leave comments. The output of the four regression models that were run shows that when compared to Twitter, Instagram has a higher beneficial impact on users’
activity on the companies’ fan pages. Besides, Instagram was found to stimulate users’
interests more by receiving more likes, comments, and more positive comments.
Moving on to H2 and H3, the moderating effect of the product type and the category on the direct relationship between different types of content and customer engagement outcomes is researched. It was expected that informative and/or persuasive content for utilitarian products would generate a more substantial and more positive impact on customer engagement
outcomes compared to entertaining content. However, the interaction effects of informative content with utilitarian values and persuasive content with utilitarian values demonstrate no significant link to customers’ liking and commenting behaviour. On top of that, no support can be found for this effect to be stronger for posts from firms operating within the electronics category compared to the travel and the fashion category. In the dataset, no situation was found where posts contained informative or persuasive content with utilitarian values for the firm operating within the fashion category, meaning that no correlations were found, and thus the variables were deleted from the analyses.
Likewise, the current study cannot provide sufficient evidence to the assumption that for hedonic products, entertaining content generates a stronger and (more) positive impact on customer engagement than informative and/or persuasive content. However, significant support can be found for this effect to be stronger for posts placed by firms operating in the fashion category compared to the travel and the electronics category for the outcome variable ratio comments/followers. On the other hand, the interaction variable Hedonic x Entertaining
51 x Fashion shows negative results on the other outcome variables (sentiment score, ratio
likes/followers, and engagement rate (%)). Thus, for firms operating in the fashion category, posts containing entertaining content and hedonic values stimulate commenting, meaning that commenting is a more popular fan page users’ behaviour than liking. In addition to that, comments are more negatively loaded. However, those negative effects aren’t significant and thus not supported. In other words, H3 could partly be supported by the outcomes of the regression models that were run.
Table 11
Overview of the (non) supported hypotheses and the corresponding dependent variables
Dependent Variable Supported
H1: Entertaining x Instagram Ratio likes/followers No
Ratio comments/followers No
Sentiment score No
Engagement rate (%) No
H2: Informative and/or persuasive x Utilitarian Ratio likes/followers No Ratio comments/followers No
Sentiment score No
Engagement rate (%) No
H2: Informative and/or persuasive x Utilitarian x Electronics Ratio likes/followers No Ratio comments/followers No
Sentiment score No
Engagement rate (%) No
H3: Entertaining x Hedonic Ratio likes/followers No
Ratio comments/followers No
Sentiment score No
Engagement rate (%) No
H3: Entertaining x Hedonic x Fashion Ratio likes/followers No
Ratio comments/followers Yes
Sentiment score No
Engagement rate (%) No
The strong support for the main effects of the moderator variable type of category on the outcome variables is worth mentioning. The travel category shows a significantly stronger and
52 positive effect on liking behaviour compared to the other industries. Also, evidence shows that content posted by firms operating in the travel category generates higher outcomes on the variable engagement rate (%) compared to other industries. In reverse, a negative effect was found for commenting behaviour on posts of firms operating within the fashion category compared to other industries. Last but not least, the sentiment score of posts placed by firms operating in the fashion category is significantly the highest, meaning that users comment more positively on content posted for products and brands within the fashion category than the travel and the electronics category. Likewise, content posted for products and brands within the electronics category results in more negatively loaded comments. Thus, the findings provide sufficient evidence that postings by firms operating in the fashion category attract more positive comments resulting in a higher sentiment score compared to the travel and the electronics category. In contrast, it threatens the ratio comments/followers. The same findings reveal that posts placed by firms operating in the travel category attract more likes and sequential a higher customer engagement rate (%) than the fashion and electronics category. It can be noticed here that there is significant evidence for a negative correlation between the sentiment score and the total amount of comments of postings placed by firms operating in the fashion category, which is in line with a previous finding of a study by Shahbaznezhad et al. (2021). The results of this paper draw further upon their findings by showing differences in this correlation for firms operating in specific categories as the significant main effects of the travel category on engagement behaviour do not show a negative correlation between sentiment score and the ratio comments/followers.
53