Kudos for CEOs – A Content Analysis on Speech Acts in CEO Posts on LinkedIn

(1)

KUDO FOR CEOs

A CONTENT ANALYSIS ON SPEECH ACTS IN CEO POSTS ON LINKEDIN

Nina Gray

11364629

Master’s Thesis

Graduate School of Communication, University of Amsterdam

Master’s programme Communication Science

Jeroen Jonkman

(2)

Abstract

LinkedIn is a professional social medium that provides its users and organizations an online platform to connect, interact and post their opinions, thoughts and professional updates. This study specifically looks at posts written by CEOs and analyzes the speech acts used in these posts and their effect on the reader’s reactions. More specifically, a content analysis on 550 LinkedIn posts written by 44 different CEOs was conducted, analyzing the presence of the different speech acts, assertive, directive, commissive, expressive and declarative, derived from the Speech Act Theory. Additionally, the number of LinkedIn’s provided kudo type reactions; like, celebrate, love, curious and insightful were also measured per post. The study further analyzes the effect of the moderators, sentiment and thought leadership, while controlling for the variables post length and followers. The results showed that different speech acts had varying effects on the individual kudo types. However, in general, the direct effects of assertive and declarative speech acts significantly increase the number of reactions of most kudo types per post. The directive speech act, on the other hand, reduces the number of kudo type reactions per post compared to the post not using the directive speech act. Negative sentiment was found to reduce the number of reactions, whereas positive sentiment showed mixed effects. Thought leadership also resulted in mixed findings. Overall, this study recommends CEOs to post on LinkedIn using any speech act in comparison to not posting at all; however, a CEO should actively think about the intent of the post as well as the desired reactions from the reader before posting.

(3)

KUDO FOR CEOs

A CONTENT ANALYSIS ON SPEECH ACTS IN CEO POSTS ON LINKEDIN

“Choose your words carefully. The potential for words to change minds and influence people, both positively and negatively, has never been stronger.” (McDermott, n.d.)

Social media provides a network of connections between users around the world (Cheung, Lee &, Rabjohn, 2008; Rooderkerk & Pauwels, 2016). The line between a user’s online and offline world is slowly blurring as people increasingly share their offline experiences in their online profiles (Boullianne, 2015; Humphreys, Gill &, Krishnamurthy, 2013). Whereas social media initially focused on personal connections, a more professional social medium has established itself (Rooderkerk & Pauwels, 2016). LinkedIn is the biggest and most popular professional social medium (Rooderkerk & Pauwels, 2016) which allows users to create a profile similar to one’s CV, build private and professional connections as well as write and share posts with other LinkedIn users. If a user chooses to post a message, their activity will be displayed on their professional profile and every person who follows that user will be able to see the post on their individual frontpage. The posting function is used both by organizations and individuals. However, it is especially important to understand how employees in highly successful and influential positions, such as chief executive officers (CEOs), make use of the posting function. As their position is commonly very high profile and associated with the success of the

organization, their online involvement is an important factor for the organization’s reputation as well as innovativeness (Finkelstein, Hambrick &, Cannella, 2009; Graffin, Boivie & Carpenter,

(4)

2012; Kesner & Sebora, 1994, Men, Tsai, Chen & Ji, 2018).

Delving into this topic, this study investigates CEO posts on LinkedIn by measuring what speech acts are used within the post and what kind of reactions/kudo they trigger among the readers. Moreover, the study also analyzes the effect of the moderators, sentiment and thought leadership in order to understand how the main relationship is impacted by the ideas and evaluations of the CEO (Stieglitz & Dang-Xuan, 2013).

The Speech Act Theory (SAT) focuses on people’s utterances and can be identified in every form of communication (i.e. spoken or written) (Akinwotu, 2013; Ordenes, Grewal, Ludwig, Ruyter, Mahr & Wetzels, 2018; Searle, 1969). More specifically, it divides people’s speech use into “performative and constative utterance” and categorizes the theory into three subdivisions (Akinwotu, 2013, p.45). First, the locutionary acts focus on the action of stating something. Second, the illocutionary acts specify the intentions in the statement. Third, the perlocutionary acts focus on the understanding of the readers (Akinwotu, 2013). This study specifically investigates the use of the second speech act, illocutionary act, by analyzing five sub-components specified by Searle’s (1996; 1976) modification of the initial understanding of the SAT. The subcomponents imply that a message’s intent can either be assertive, directive, commissive, expressive or declarative (Akinwotu, 2013; Ordenes et al., 2018; Searle, 1976; Simon & Dejica-Cartis, 2015).

As messages have different intentions, they may consequently affect the readers

differently. LinkedIn allows for reactions beyond liking a post, such as choosing to celebrate or love the post or finding it curious or insightful. However, unlike reaction types on media like Facebook (like, love, haha, wow, sad, angry), LinkedIn exclusively provides positive reactions (like, celebrate, love, curious and insightful), referred to as kudo types in this study. The term

(5)

kudo, which originates from Greek meaning “praise”, was chosen because the readers only have the option to react with different positive forms of interest (Hench, 1963). Therefore, this

research analyzes the effect of the five SAT acts on these five specific kudo types.

This study contributes insights into the various user-reactions on LinkedIn, which have not yet been investigated in any other study. Therefore, this research is theoretically relevant as it provides insight to a communication feature which is not yet theoretically established and

contributes to the increasingly more researched field of social media communication between organizations and their stakeholders (Saxton & Waters, 2014). Especially by focusing the study on LinkedIn, theoretical insight is created on a platform which substantially differentiates itself from the more classic social and unprofessional platforms such as Facebook (Rooderkerk & Pauwels, 2016).

Concerning the social relevance, this study aims to contribute to the understanding of message optimization. Unlike Facebook, where the presence of organizations is merely tolerated by users like millennials (Saxton & Waters, 2014), LinkedIn is specifically designed to fill this niche and provide a professional social medium. Therefore, this study provides guidance for users on how their posting activity on LinkedIn can be optimized to achieve greater reactions. This may generate more attention to people’s posts and may also encourage more users to post on LinkedIn. Moreover, this study is especially valuable to public relation professionals as they may support the CEO’s social media management.

For this reason, this study aims to understand, how users react to different types of speech acts by investigating the following research question:

(6)

effect on the kudo types; and how, and to what extent, are these relationships moderated by the post’s sentiment and thought leadership expressed by the CEO?

Theoretical Background

Speech Act Theory

The Speech Act Theory divides linguistic acts into three categories (Searle, 1996; Goldspink, 2010). Yet, the theory is frequently mistaken as representing merely the second act, i.e. illocutionary acts (Akinwotu, 2013). The illocutionary acts can be analyzed in two forms; the literal and the pragmatic meaning (Goldspink, 2010). By looking at the literal form a direct, dictionary meaning of the word is implied, whereas the pragmatic form requires context and intent of the speaker (Goldspink, 2010; Stiles, 1992). This is a crucial as it emphasizes the importance of the speaker as well as the reader. As the reader is able to “syntactical(ly) encode emotions”, readers can make sense of the context and meaning behind the message of the post (Stieglitz & Dan-Xuan, 2013, p. 223). This implies that there is a relation between the individual SAT acts coded and the different kudo types. As the number of kudo types per post cannot fall below zero, the effect of the SAT acts can only be evaluated as more or less positive than if the speech act were not used.

Speech Act Theory & Kudo Types

The effect of messages on different kudo types has been analyzed in many studies by measuring reactions on Facebook posts and comments (Larsson, 2016; Kaur, Balakrishnan, Rana, &

(7)

Sinniah, 2019). The possibility to select different kudo types is also available on LinkedIn and thus the concepts of Facebook reactions can arguably also be depicted in the context of LinkedIn posts.

Concerning the specific speech acts, it was found that assertive and expressive acts are considered to be more easily processed by the readers and are also more frequently used in posts (Ordenes et al., 2018; Carr et al. 2012). Conversely, directive acts in posts were expected to have less reactions and shares as the required action is considered “conclusive” and reduces the possibility for discussions or interpretation (Ordenes et al., 2018, p.991). Moreover, the speech acts commissive and declarative are least frequently used in posts as they are conveyed more through the content of what is being said, i.e. by announcing/declaring, and are less distinctive categories assertive, declarative and expressive (Ordenes et al., 2018).

Regarding the specific kudo types, like is not merely an expression of agreement or positive emotion (Turnbull & Jenkins, 2016) but is also the default reaction on LinkedIn.

Therefore, the like reaction on Facebook is still being used more frequently and is in terms of the total number of reactions superior to the other reactions (Larsson, 2016). Gaining a high number of likes is desirable because it suggests more agreement amongst the readers/followers and thus a higher popularity of the post (Swani et al., 2017; Turnbull & Jenkins, 2016). The desire to

positively influence others’ impression of oneself has been outlined in Goffman’s

self-presentation theory (Bareket-Bojmel, Moran & Shahar, 2016; Goffman, 1956). When applied to the online environment, the theory implies that people utilize social media to create impressions of themselves which are typically better than their real personality (Goffman, 1969; Qi, Mondo, Fang & Deng, 2018; Schwartz & Halegoua, 2015). The theory further suggests that social media users see themselves as leaders of their readers/audience/followers and “tend to make others act

(8)

in accordance with [their] plan” (Qi et al., 2018, p.97). This suggests that users such as CEOs, actively want to encourage the readers to think or act according to their ideas and positively assess the CEO. This implies that any speech act can be used to make the CEO’s impression more positive. For example, the reader can be told to watch a promotional video (directive), the CEO may express their opinion against sexism at the workplace (expressive), etc. However, as it has been found in a study on self-presentation on Facebook, some user’s go as far as to create a fake-self in order to increase their image (Bareket-Bojmel et al., 2016; Gil-Or, Levi-Belz, & Turel, 2015). As the assertive speech act requires the information to be truthful and also be perceived as such, it suggests that posts featuring assertive speech acts may not be as efficient in gaining more likes and improving CEOs’ impressions. Hence, the following hypothesis is proposed:

H1: Like is affected more positively by the speech acts directive, commissive, expressive and declarative than by the assertive speech act.

Celebrate and love

Similar to like, the kudo types celebrate and love, according to their literal meaning can be understood as expressions of positive emotions and enthusiasm. Studies including Kaur et al. (2019) on Facebook user reactions in health-related forums, have shown that users are more inclined to express positive reactions, especially when wanting to show support (Frison & Eggermont, 2015; Oh, Rao & Agrawal, 2013). The study claimed that 70% of all Facebook reactions were positive reactions, of which 41% of the users expressed love, 26% expressed wow and 3% reacted with haha. Even the sad reactions were shown to be used to express support for other members (Kaur et al., 2019). Such behavior is in alignment with the concepts in the

(9)

selective exposure theory, as humans are believed to actively avoid information which they disagree with or find uncomfortable (Bennett & Iyengar, 2008; Mutz, 2006). However, according to Larsson (2016), Facebook users are more likely to engage through actions like sharing if the content is negative or upsetting. The contradictions may be related to the content of the posts in each study due to emotional contagion. Emotional contagion is “the spread of mood and affect through populations by simple exposure” which has been suggested to influence the readers emotional state and can be transferred through social media by close connections of the reader, but also through leadership situations (Heath, 1996; Larson & Almeida, 1999; Stieglitz & Dang-Xuan, 2013, p.222; Sy, Côté & Saavedra, 2005). For this reason, it is hypothesized that a more emotionally stressing speech act will trigger the use of more emotionally related kudo types:

H2a: Love is affected more positively by the expressive speech act compared to the other speech acts (assertive, commissive, declarative, and directive).

H2b: Celebrate is affected more positively by the expressive speech act compared to the other speech acts (assertive, commissive, declarative, and directive).

Curious and insightful

Unlike like, love and celebrate, the kudo types curious and insightful are perceived in this study as more information based. These two kudo types strongly differ from the emotion-based reactions provided by Facebook and may emphasize the professional foundation of LinkedIn (Rooderkerk & Pauwels, 2016). Concerning curiosity, a study investigating a five-dimensional curiosity scale describes curiosity as the eagerness for new information (Kashdan, Stiksma, Disabato, Mcknight, Bekier, Kaji & Lazarus, 2018). The study identified four types of curious

(10)

people, the Fascinated, Problem-Solvers, Empathizers, and Avoiders, and showed in the study that the Fascinated and the Emphasizer displayed an overall higher interaction on various social media (Kashdan et al., 2018). A reason for the higher interaction of curious, information-seeking users may be linked to the uses and gratification theory, which looks at the reasons why people use certain media (Griffin, 2012). As LinkedIn is a very professionally specialized social

network, users who actively look at their newsfeed and get exposed to the CEO’s posts may also choose this social network site because of its business content and updates. Consequently, in alignment with the theory, it suggests that users desire information on the CEO and the

organization, and therefore, may quench their cognitive needs for acquiring knowledge (Griffin, 2012). As the foundation of curious and insightful is inherently a reaction to being given

information, it can be assumed that speech acts which are more informative such as assertive, commissive and declarative may create a more positive reaction. For this reason, the following hypotheses are proposed:

H3a: Curious is affected more positively by the speech acts assertive, commissive and declarative compared to the expressive and directive speech acts.

H3b: Insightful is affected more positively by the speech acts assertive, commissive and declarative compared to the expressive and directive speech acts.

SAT acts & Thought Leadership

According to Rooderkerk and Pauwels (2016) study on online discussion forums such as LinkedIn, organizations aim to be thought leaders. To establish such a position, a platform must provide the possibility to communicate novel thoughts to the public (van Dijk, 2013). More

(11)

specifically, in Gibbins-Klein’s (2011) article, she outlines four elements of (REAL) thought leadership; Reach, Engagement, Authority and Longevity. When applying these principles to the position of a CEO on LinkedIn, the social media platform allows for CEOs to reach their

audience and be recognized. Engagement is also high as CEOs can directly communicate with the readers through posts and thus engage the audience. Authority is instantly satisfied with the ranking of the position and most importantly, the ideas need to resonate in the audience’s mind long after reading the CEO’s message, hence requiring intriguing and challenging ideas. By presenting novel ideas or concepts, the CEO has the opportunity to trigger discussion with the readers. Yet, actual engagement can only be achieved if the readers respond to the ideas (Rooderkerk & Pauwels, 2016). Moreover, engaging a response from the readers, has been described as part of the thought leadership. Barry and Gironda (2019) included “provoking new mindsets for addressing upcoming challenges” and “drive conversations around shared passions” as functions of the thought leadership (p.141). For this reason, the presence of thought

leadership, if combined with the adjacent speech acts, may emphasize the understanding and motivation of the readers.

Additionally, literature on knowledge construction emphasizes that LinkedIn is a successful platform for sharing thought leading posts, as LinkedIn is a social network site characterized as trusting and often provides a sense of belonging as well as a place for

cooperation and knowledge articulation (Li, Cox & Wang., 2018). Therefore, if CEOs use their own account for thought-leading messages, they could potentially increase the benefits of their message and encourage engagement and reactions. More specifically, since thought leadership explicitly encourages the expression of novel ideas it may trigger the reader to actively think about the post and increase reactions of curiosity and insightfulness.

(12)

H4: Thought leadership more positively moderates the main relationship between the SAT acts (assertive, directive, expressive, commissive and declarative) and the kudo types

curious and insightful compared to the relationship between all SAT acts and the kudo types like, love and celebrate.

SAT acts & Sentiment

Research on the use of emotions and sentiment has demonstrated that the use of emotional appeals significantly influences the audience and may therefore function as an efficient and persuasive tool (Stieglitz & Dang-Xuan, 2013). Furthermore, the readers of messages through computer mediated communication are capable of understanding the emotions expressed in the post by identifying not only verbal cues but also paralinguistic cues (Stieglitz & Dan-Xuan, 2013), which in turn, may influence the perception of the audience toward the post. Additionally, the social presence theory suggests that different media can influence the way we perceive a sense of closeness or togetherness with other people (Men et al., 2018). Concerning LinkedIn, it allows for the expression of social clues through writing a message in a single post. LinkedIn may also provide a closer sense of presence compared to other media like TV or press releases, as it allows for interaction from the reader’s side such as liking, commenting and sharing. By expressing sentiment, the followers, therefore, may feel more connected to the CEO, which suggests that readers can also relate to the authoritative figures better and may be more inclined to agree with them by liking the post (Turnbull & Jenkins, 2016).

According to Joyce and Kraut (2006), positive communication may encourage participation by the readers (Stieglitz & Dan-Xuan, 2013). Using sentiment positively to

(13)

establish open, personal and intimate communication was further found to increase the quality of the relationship with the public, as CEOs, that are perceived as social and authentic, are more quickly viewed as real people (Huy & Shiilov, 2012; Men & Tsai, 2016). However, negative sentiment was also found to be associated with higher reactions on social media (Stieglitz & Dan-Xuan, 2013). Especially, on platforms such as Twitter, negative sentiment was found to increase the amount of comments and the likelihood of virality of the posts which shows that also negative sentiment engages the readers (Stieglitz & Dan-Xuan, 2013). Moreover, both positive and negative sentiment are proposed to increase the readers’ attention (Stieglitz & Dan-Xuan, 2013), with studies showing that online contagion occur for both positive and negative sentiment (Bareket-Bojmel, Moran & Shahar, 2016). This suggests that within this study, the contagion of the positive sentiment will more positively affect the use of positive emotional kudo types. Consequently, the following hypotheses are proposed:

H5a: Positive sentiment more positively moderates the main relationship between the SAT acts (assertive, directive, expressive, commissive and declarative) and the kudo types like, celebrate and love compared to the relationship between all SAT acts and the kudo types curious and insightful.

H5b: Negative sentiment less positively moderates the main relationship between all SAT acts (assertive, directive, expressive, commissive and declarative) and the kudo types like,

celebrate and love compared to the relationship between all SAT acts and the kudo types curious and insightful.

(14)

This study has conducted a manual quantitative content analysis on LinkedIn posts written by users who work as Chief Executive Officers (CEO). All posts were posted between 2017-2020. For a post to be coded, the length of the text had to be minimally one sentence. The minimum requirement ensures that the measured speech acts are sufficiently represented within the post and the CEO has expressed some sort of message. Posts were excluded if they only hashtags or are a repost of another article with no extra text from the CEO.

The posts coded are derived from an umbrella project on CEO LinkedIn posts, N=2024 posts by 128 CEOs1_{. This study coded n =550 posts, written by 44 CEOs. The sample was} selected based on a convenience sample and the selection was based on the post’s qualification and the type of company. Only companies from financial, technological and software industries were chosen (e.g. Siemens, PayPal, etc.) in order to compare companies whose product is numeric instead of communication based. Moreover, according to a study by Distaso, Mccorkindale and Agugliaro (2015) the technology industry had the second highest “social media best practice score” after FMCG and Financial Services. These scored best on social media like YouTube which allows finance companies to appear more “humanized” compared to platforms like Facebook. As this study investigates posts written by CEOs, emphasis falls on trust and quality interactions which appear to be satisfied by the social media action of the chosen industries.

Variables

The study investigates four main variable groups; the SAT acts, the kudo types and the

moderators, sentiment and thought leadership. The SAT acts are the independent variables which potentially affect the dependent variables, kudo types. Furthermore, the variables sentiment and

(15)

thought leadership are hypothesized as moderators between the SAT acts and the kudo types of the post.

Speech Act Variables. The variable SAT acts represent the five sub-components of the illocutionary speech act theory; assertive, commissive, directive, expressive and declarative. More specifically, assertive, is measured by the presence of a factual, non-opinionated statement which the speaker/CEO believes to be true. The statement may inform, confirm or state

something (Simon & Dejica-Cartis, 2015; Akinwotu, 2013). Directive acts aim to create a reaction from the readers, emotional or behavioral (Akinwotu, 2013; Searle 1969;1976). Indications for a directive act may be the CEO challenging, demanding or encouraging the readers (Simon & Dejica-Cartis, 2015; Akinwotu, 2013). However, unless an act was explicitly stated such as: “Watch our video!” or “Keep in mind…”, the directive act was not coded as present. Commissive acts imply that the content of the post commits the readers to something that will happen in the future (Akinwotu, 2013; Searle 1969;1976). Common phrases associated with commissive acts are promises, offers, or even threats that are not happening at the moment of the post or prior to the posting. Expressive acts are focused on the expression of emotion by using language which reveals how the CEO feels or what his/her opinion towards a specific topic is (Akinwotu, 2013; Searle 1969;1976; Simon & Dejica-Cartis, 2015). Lastly, declarative acts focus on declaring a change happening currently. Change is a key term as CEOs posting about their current activity was not included in the coding; however, statements such as resigning, announcing or naming are common examples (Akinwotu, 2013; Searle 1969;1976; Simon & Dejica-Cartis, 2015). More examples and explanations can be found in the Appendix (Q).

(16)

Kudo Type. The kudo types are the multiple options a LinkedIn user can choose when reacting to a post. Unlike platforms such as Facebook, LinkedIn’s different reaction options are less emotionally extreme but rather allows the users to select the following options: like,

celebrate, love, curious and insightful. This study interprets the kudo types based on the literal meaning of the words, hence like indicating that the readers enjoyed/agreed with the content, celebrate indicates that readers praise or commemorate the content, love suggests that they have strong positive emotions toward the post, curious suggests eagerness to learn from the post and insightful indicates that the post shows perceptive and information. A LinkedIn user can select a kudo type by placing the cursor above the “like” button under each post. If no kudo is chosen, the kudo type “like” is a default button.

The study has two moderating variables, sentiment and thought leadership. As the two moderating variables have been coded in the umbrella project, the values in this study are used exactly as in the umbrella project1_.

Sentiment. Within this study, the focus lies on the positive sentiment and the negative sentiment in order to gain clearer insight into which exact type of sentiment is moderating the relationship. Therefore, two dummy variables were created, coding both types of sentiment 1 for present and 0 for not present. According to the umbrella project’s codebook1 _{the presence of} positive sentiment meant that content included “describes an adequate solution, needs to be constructive, mentions positive trends/events, conveys optimism about the future/topic/self”1. Negative sentiment was described as content “focused on problems/issues, critical towards solutions, pessimistic about the future/topic/self, mentions negative trends/even”1. With each type of sentiment, interaction variables were computed by multiplying, for example, positive sentiment with each individual SAT act.

(17)

Thought leadership. Thought leadership is a dichotomous variable measuring whether the post is thought leading or thought following. If the post included proactive content about what the CEO, the company or the readers will/should do about an event or topic, then the post was marked with a 1 for thought leadership1. If the content was reactive, the post was coded with 0 for thought followership1. Again, an interaction variable was created with thought leadership and each kudo type to check for an interaction effect.

Reliability

To ensure more reliable data the newly coded variables have undergone several

intercoder reliability tests. During the process of writing the codebook, the data was repeatedly tested in small samples with a second coder to ensure a more reliable codebook. Once the codebook was completed, 55 posts were coded. The inter-coder reliability test was run three times in total by three different coders as the first two had shown unsatisfactory Krippendorff alpha scores. A suspected reason for the low scores is because the inter-coder training for the first coder was conducted two weeks prior to the actual coding, suggesting that key aspects of the training were not refreshed sufficiently. The second partially unreliable scores emphasized the issues with interpretation of the speech acts. For this reason, the third coder underwent more extensive training, where key words for each speech act were listed, the codebook was read and additional literature on speech acts were discussed. This reduced issues of subjectivity and allowed for a deeper understanding of the speech acts’ origins. A total of 50 codes were double coded with the following ICR scores; assertive had a Krippendorff's alpha of .68 and a percentile agreement of 90%, directive had an alpha of .87 and a PA of 94%, commissive had an alpha of .73 and a PA of 96%, expressive had an alpha of .76 and a PA of 90% and lastly, declarative had

(18)

an alpha of .67 with a PA of 86%. For the scores of the first two reliability tests see Appendix A. Concerning the moderating variables, the umbrella study used the lotus measure. Sentiment had a 74% PA and a lotus of .89, thought leadership also had a 74% PA and a lotus of .69.

Control variables

Throughout the statistical analyses, the continuous variables post length and followers were included as control variables. The post length was added as a control variable because more speech acts may occur in texts which are longer. Followers was the number of followers which the CEO had during the moment of coding. As higher follower counts may lead to higher exposure, this effect had to be accounted for.

Statistical analyses

Negative binomial regression analyses were used to investigate the hypotheses. Due to the natural distribution of reactions on social media, a large number of posts with lesser reactions were found and only few posts with a high reaction were coded resulting in a highly, positively skewed data distribution (Saxton & Waters, 2014; Trilling, Tolochko & Burscher, 2017). Per kudo type, two negative binomial regressions were conducted. First, an analysis of the direct relationship between all speech acts and the dependent variable was done. Second, a regression including all speech acts and all interaction variables with the dependent variable was conducted. A total of 15 regressions were completed (Appendix B-P). The incidence rate ratios (IRR) were measured for each regression and were recorded by evaluating any value below 1 as a negative effect and above 1 as a positive effect on the dependent variable (Jonkman et al., 2020). More specifically, if a regression has an IRR of .86 it implies that a change of one-unit results in 86%

(19)

expected reactions. However, if the IRR is 1.7 then a one-unit increase suggests 170% expected reactions (Trilling, Tolochko & Burscher, 2017). However, since it is impossible for a post to have negative likes, an IRR under 1 implies that the post would get less positive reactions, meaning less reaction than if the speech act were not used in the post at all.

Results

This analysis investigates the relationship between the five illocutionary speech acts on the different kudo types, and the role of moderators, sentiment and thought leadership. The variable post length and follower count are kept as control variables. The effect of the SAT acts, assertive (M = .25, SD = .43), directive (M = .22, SD = .42), commissive (M = .08, SD = .27), expressive (M = .71, SD = .452) and declarative (M = .23, SD = .42) showed mixed results with the different kudo types, likes (M = 851.26, SD = 1935.88), celebrate (M = 19.61, SD = 62.1), love (M = 11.93, SD = 49.07), curious (M = .80, SD = 3.53) and insightful (M = 2.31, SD = 19.86). The moderators had an average of M = .03 for positive sentiment (SD = .16), M = .55 for

negative sentiment (SD = .50) and M = .26 for thought leadership (SD = .44). Lastly, the control variables post length (M = 396.17, SD = 268.25) and followers (M = 375986.59, SD = 0) were controlled for in all analyses.

All SAT act variables had a minimum of zero and a maximum value of one. The most commonly selected speech act was expressive which was used in 393 posts. Followed by assertive, declarative and directive acts which appeared in 126, 125 and 122 of all posts,

respectively. The least frequently used speech act in all posts was commissive act which was only used 42 times in all coded posts.

(20)

Regarding the kudo types, the maximum amount of reactions per post was 18,741 likes, 824 celebrates, 521 loves, 48 curious and 446 insightful reactions. The average post had approximately 851 likes, 20 celebrate, 12 love reactions, 1 curious and 2 insightful reactions.

SAT acts & Kudo Types

Since every post received reactions, effects can only be less positive or more positive than if the speech act was not used in the post. The results from the Negative Binomial Regression of SAT acts on each individual kudo type showed significant results for the kudo types, like, celebrate, love and insightful. No significant direct effect on curious was found. Therefore, the third hypothesis on the effects of the SAT acts on curious (H3a) can neither be supported nor refuted. All tables showing the output of the regression can be found in the Appendix (B-P).

Concerning the significant direct effects, like was only found to be less positively affected by the directive speech act, (IRR = .73, p = .019). As the directive speech act was hypothesized to more positively affect like, but has an IRR below 1, the first hypothesis is not supported (H1). Celebrate was the most affected variable, as the speech acts assertive, directive and declarative were found to have a direct effect. Assertive and declarative were found to have a more positive effect on the choice to select celebrate with (IRR = 1.58, p = .017) and (IRR = 2.21, p < .001), respectively. Directive act was again found to have a less positive effect on the kudo type (IRR = .42, p < .001). As assertive did have a more positive effect but declarative did as well, the second hypothesis on celebrate is not supported (H2a). Regarding love, findings showed that directive acts less positively affected the kudo types, love, (IRR = .35, p < .001). While the less positive effect is in alignment with the hypothesis (H2b), the effects of the other

(21)

speech acts were insignificant. Thus, the second hypothesis can only be partially supported (H2b). Lastly, the effect of the kudo type on insightful, only had a more positive significant effect of assertive on insightful, (IRR = 1.8, p = .028) which was in accordance with the hypothesis (H3b). However, again, as no other direct effects were found, hence the hypothesis is partially supported (H3b). .

SAT acts, Kudo Types & Moderators

Thought Leadership. Regarding thought leadership as a moderator, the interaction effect with individual SAT acts showed mixed effects. The interaction effect on likes showed that if a post uses expressive acts and thought leadership it will less positively affect the number of likes which the post receives than if it were not using these two variables, (IRR = .71, p = .034). Concerning the interaction effect on celebrate, the use of directive acts strongly, more positively affects the selection of celebrate when moderated by thought leadership, (IRR = 5.26, p < .001). However, when declarative acts are used with thought leadership moderating then the amount of celebrate reactions will be less than if the interaction was not present in the post, (IRR = .44, p = .028). The effect on love, was also affected by the interaction between thought leadership, directive and declarative and similarly showed that the interaction with directive speech acts increased the likelihood for users to choose love when exposed to the post, (IRR = .5.67, p < .001) but less positively affected the choice, love, when declarative and thought leadership was used (IRR = .3, p = .003). Concerning curious, the moderator, thought leadership, more positively affected the effect of directive and expressive acts on curious, (IRR = 3.46, p = .029) and (IRR = 2.96, p < .001), respectively. Lastly, insightful was not moderated by thought

(22)

leadership when using a confidence interval of 95%.

As curious and insightful were the only kudo types hypothesized to be more positively affected by the moderator, thought leadership, but more positive moderation was also found on love and celebrate, the fourth hypothesis can be refuted (H4).

Sentiment. When including the moderating variable, sentiment, both types of the variables, positive sentiment and negative sentiment, showed moderating effects on the relationship between SAT acts and kudo types.

The interaction effect on love showed more positive effects when interacting with the speech acts assertive and directive, resulting in a much higher positive choice for love compared to posts which did not feature these interactions, (IRR = 2.92, p = .009) and (IRR = 3.82, p = .002), respectively. Moreover, the interaction of positive sentiment with commissive and with expressive acts concluded significant effects on insightful. Commissive when interacting with positive sentiment showed the highest positive effect of all regressions in this study, indicating that the use of commissive and positive sentiment much more positively affects the number of insightful reactions, (IRR = 8.6 p = .043). Conversely, the interaction between expressive and positive sentiment less positively affects insightful, (IRR = .22, p = .025). Based on the very mixed results, especially concerning insightful which was hypothesized to be less positively affected than the kudo types, like, love and celebrate, the hypothesis is not supported (H5a).

Regarding the moderation of negative sentiment, the only significant moderation found was the notably, less positively affected like by posts which used both commissive speech acts and negative sentiment, (IRR = .05, p = .038). This finding strongly aligns with the hypothesis; however, due to nonsignificant results, the hypothesis is only partially supported (H5b).

(23)

Conclusion

This study aims to shed further light on the success of different types of speech acts used by CEOs on LinkedIn with the intent of each message being divided into five speech acts, which are each tested for their likelihood to affect the amount kudo types for each post. Additionally, the study measured whether sentiment and thought leadership affect the relationship between the SAT acts and the kudo types used in the post.

When more generally addressing the effect of SAT acts directly on the different kudo types, results showed that overall assertive and declarative acts had the most positive effects on kudo types, in particular on celebrate and insightful. The success of assertive and declarative acts may align with the type of content which is announced. Since both speech acts more positively affected celebrate, it suggests that a substantial amount of the declarative and assertive content may have been positive or encouraging. This observation further aligns with the study by Kaur et al.’s (2019) who claim that users prefer reacting positively and are therefore potentially more inclined to react to positive posts in general. Directive, however, showed only less positive direct effects specifically on like, celebrate and love. As all significant effects are less positive, it implies a consistent issue with the speech act. While other speech acts require one’s attention, this act requests action (Ordenes et al., 2018) which may distract the readers from reacting to the post.

Addressing the specific hypotheses, the reason for the rejected first hypothesis may be because like is not just a kudo type but also the default button. Whereas all other kudo types require users to actively select them, like is the “original” reaction and is also believed to still be

(24)

the most popular (Larsson, 2017). This justification is supported by the overwhelmingly higher average number of likes per post compared to all other kudo types combined. Therefore, like may not be used in response to a specific speech act but just as a more general reaction.

Concerning the conclusions of the second hypotheses regarding the direct effect on love and celebrate, an overarching issue may be linked to the insignificance of the expressive act. Expressive did not only represent communicating one’s feelings such as being happy or sad but also included the expression of opinions. Similar to the concept of emotional contagion, readers may perceive the intent of a post differently by judging it more as a fact than an opinion

(Stieglitz & Dang-Xuan, 2013). Moreover, while this study aimed to only code literal

expressions of emotions, much of communication is also the tone of a message which may have affected the choice of kudo types by the readers. This suggests that readers may not only react to the CEO’s emotions but may also make active choices based on the form in which the emotion is conveyed. Moreover, celebrate was also found to be more frequently chosen as a reaction when the CEO used assertive and declarative speech acts. This is insightful because it shows that readers also follow the CEO for factual information and updates on the CEO’s or company’s situation. This implies that the celebrate reaction may be utilized similarly to the more information-based kudo types, curious and insightful, and may be interpreted as a reaction to information, rather than just positive contagion.

The finding of the effect on insightful supports the third hypothesis as assertive was found to affect insightful more positively. Especially since assertive is the only speech act affecting insightful, it emphasizes the relation between the presence of factual information and readers finding it insightful. Moreover, this supports the idea that LinkedIn users actively use this social medium for cognitive needs (Griffin, 2012). Whereas informing/updating readers, is only

(25)

one function of the website, it emphasizes the purpose of LinkedIn as a source for business news and information and places LinkedIn as a specific source for users to satisfy their cognitive needs (Griffin, 2012).

Beyond the direct effect, the diverse effect of the moderators is crucial within this study as all moderators showed significant moderating effects.

Interactions with sentiment, both positive and negative, combined showed to affect all kudo types with the exception of curious. Such frequent interaction implies that the stance of the CEO actively impacts the relationship between the SAT acts as well as the way readers react to the post.

Concerning the negative sentiment, the interaction with commissive acts strongly reduced the number of likes per post. This less positive effect from an interaction with negative sentiment follows the findings in Kaur et al.’s (2019) study, as users were found to prefer reacting to positive information more than to negative.

Positive sentiment, on the other hand, showed more mixed interactions. First, positive sentiment when interacting with assertive and directive acts had a more positive effect on the selection of love. The interaction with positive sentiment appeared to stimulate the post toward a more emotional context. As readers are found to understand emotions embedded in the context, using positive sentiment with the use of assertive or directive acts may still stimulate a greater number of happier reactions like love (Stieglitz & Dan-Xuan, 2013; Harris & Paradice, 2007). Besides love, positive sentiment also sparked a significantly more positive effect between commissive acts and insightful. The key difference between this interaction effect compared to the other interaction effects on love is that insightful is an informational-based kudo type. As previous studies showed that sentiment can be an efficient persuasive tool and increase attention,

(26)

it suggests that readers pay more attention to the commissive post when positive sentiment is involved (Stieglitz & Dang-Xuan, 2013). Moreover, this links to the selective exposure theory as readers may be more inclined to acknowledge the announcements if they are positively interested in them and only then dedicate actual attention to the information (Bennett & Iyengar, 2008; Mutz, 2006). Interestingly, however, interaction with positive sentiment also less positively affected insightful reactions when combined with expressive acts. This interaction contradicts the theories and findings supporting positive content encouraging reactions, as positive sentiment and expressive acts would suggest a very positive post (Kaur et al., 2019, (Bennett & Iyengar, 2008; Mutz, 2006). Therefore, the less positive effect on insightful provides two implications. Firstly, it supports the insightful kudo type being used predominantly as a reaction to informative posts and, secondly, it further emphasizes the need for users to go on LinkedIn to gather

information, suggesting again that this social medium serves a specific use (Griffin, 2012). The moderator, thought leadership, affected the relationship with all kudo types except insightful by moderating the relationship with the speech acts directive, expressive and

declarative. The interaction effect between directive and thought leadership was much more positive with celebrate, love and curious. The significant interactions show that organizations which actively engage in thought leadership can also increase the amount of reactions. This is important as the directive act only had less positive effects when not moderated by thought leadership. However, unlike the interaction effect with directive, interactions with declarative predominantly generated less positive effects. This effect contradicts the explanation of

successful thought leadership tasks by Barry and Gironda (2019) which list that amongst other tasks thought leaders should be mind-challenging. A reason for this contradiction may be that users found the content overwhelming or challenging and therefore reacted less to the post.

(27)

Regarding the interaction with expressive and thought leadership, it showed a less positive effect on likes; however, a significantly more positive effect on the kudo type, curious. This moderating effect on curious is especially important, because thought leadership triggered an effect between expressive and directive acts on curious which did not exist as a direct effect. This shows that if a CEO wants readers to react more with curious when sharing their opinion or calling for action, they should also express thought leading ideas.

In conclusion, all posts received some type of kudo. Therefore, it is important to note that any post is more advantageous to the CEO than not posting at all. However, this study showed that different types of speech acts affect the kudo types differently. For that reason, it is crucial to ask two questions before posting:

1. Why am I writing this post?

2. What do I want to achieve with this post?

This study emphasized that the intention of a post is reflected in the speech acts used; however, the reactions to the different speech acts are not naturally congruent with the speech act’s intentions. A CEO must analyze the importance of the post and actively consider what different kudo types mean and which one(s) the CEO desires most to generate.

Beyond the outcome of the study, several limitations should be noted. The most

significant among them is the exposure of the posts. It is unclear how many users read the posts even when they are actively following the CEO. The number of reactions does not provide an indication as to how many of the readers have actually chosen to react. Furthermore, the line between utterance and implication of language when coding for speech acts, even after multiple coders, was still difficult to separate. The nature of the data is partially subjective, especially

(28)

when assessing variables such as the expressive or assertive acts which heavily depend on the difference between what an opinion is and what not. Lastly, a post on average had 24 time more likes than all kudo types combined which makes predictions more difficult if a CEO aims for overall higher reactions.

For this reason, this study has the following future research recommendations. As the data showed many different and possibly contradictory relationships, it suggests that future research should have more control variables such as posting time, relation to current affairs or gender. As the CEOs measured in this study were nearly all male, future research should incorporate more female CEOs to test for gender differences as well. Furthermore, this study should be compared with reactions to posts posted by non-influential or highly ranked employees in order to potentially detect effects of leadership. For example, comparing CEO posts to posts by regular employees of the same company.

In conclusion, this content analysis contributed further insight into the use of communication on the social media platform, LinkedIn and aims to emphasize the importance of speech acts and its

diverse reactions online.

References

Akinwotu, S. A. (2013). A Speech Act Analysis of the Acceptance of Nomination Speeches of Chief Obafemi Awolowo and Chief M.K.O. Abiola. English Linguistics Research, 2(1). doi: 10.5430/elr.v2n1p43

(29)

Bareket-Bojmel, L., Moran, S., & Shahar, G. (2016). Strategic self-presentation on Facebook: Personal motives and audience response to online behavior. Computers in Human Behavior, 55, 788–795. https://doi.org/10.1016/j.chb.2015.10.033

Barry, J. M., & Gironda, J. T. (2019). Operationalizing thought leadership for online B2B marketing. Industrial Marketing Management, 81, 138–159. doi:

10.1016/j.indmarman.2017.11.005

Bennett, W. L., & Iyengar, S. (2008). A New Era of Minimal Effects? The Changing

Foundations of Political Communication. Journal of Communication, 58(4), 707–731. https://doi.org/10.1111/j.1460-2466.2008.00410.x

Boullianne, S. (2015). Social media use and participation: a meta-analysis of current research. Information, Communication & Society, 18(5), 524–538. doi: 10.1080/1369118x.2015.1008542

Cheung, C. M., Lee, M. K., & Rabjohn, N. (2008). The impact of electronic word‐of‐ mouth. Internet Research, 18(3), 229–247. doi: 10.1108/10662240810883290

Distaso, M. W., Mccorkindale, T., & Agugliaro, A. (2015). America's Most Admired Companies Social Media Industry Divide. Journal of Promotion Management, 21(2), 163–189. https://doi.org/10.1080/10496491.2014.996799

Finkelstein, S., Hambrick, D., & Cannella, A. A. (2009). Strategic leadership: theory and research on executives, top management teams, and boards. Oxford Univ. Press.

(30)

Frison, E. & Eggermont, S. (2015). The impact of daily stress on adolescents’ depressed mood: The role of social support seeking through Facebook, Computers in Human Behavior. 44, 315-325.

Gibbins‐Klein, M. (2011). Winning by thinking: how to create a culture of thought leadership in your organization. Development and Learning in Organizations: An International

Journal, 25(1), 8–10. doi: 10.1108/14777281111096762

Graffin, S. D., Boivie, S., & Carpenter, M. A. (2012). Examining CEO succession and the role of heuristics in early-stage CEO evaluation. Strategic Management Journal, 34(4), 383– 403. doi: 10.1002/smj.2019

Goldspink, C. (2010). Normative behaviour in Wikipedia. Information, Communication & Society, 13(5), 652–673. doi: 10.1080/13691180903214523

Goffman, E. (1969). Strategic interaction. University of Pennsylvania Press.

Goffman, E. (1959) The presentation of self in everyday life Anchor, New York (1959)

Harris, R.B., and Paradice, D. (2007). An investigation of the computer-mediated

communication of emotion. Journal of Applied Sciences Research, 3(12), 2081–2090. Heath, C. (1996). Do people prefer to pass along good news or bad news? Valence and relevance

of news as a predictor of transmission propensity. Organizational Behavior and Human Decision Processes, 68, 2, 79–94. doi: 0749-5978/96

Hench, A. L. (1963). Singular, "Kudos" and Plural, "Kudos"; Singular, "Kudo". American Speech, 38(4), 303–305.

(31)

information. Information, Communication & Society, 17(7), 843–857. doi: 10.1080/1369118x.2013.848917

Joyce, E., & Kraut, R. E. (2006). Predicting Continued Participation in Newsgroups. Journal of Computer-Mediated Communication, 11(3), 723–747. https://doi.org/10.1111/j.1083-6101.2006.00033.x

Griffin, E, Katz, E. (2012). Uses and Gratifications. In A first look at Communication Theory (Vol. 8, pp. 357-365). New York, New York: McGraw-Hill.

Kashdan, T. B., Stiksma, M. C., Disabato, D. J., Mcknight, P. E., Bekier, J., Kaji, J., & Lazarus, R. (2018). The five-dimensional curiosity scale: Capturing the bandwidth of curiosity and identifying four unique subgroups of curious people. Journal of Research in Personality, 73, 130–149. https://doi.org/10.1016/j.jrp.2017.11.011

Kaur, W., Balakrishnan, V., Rana, O., & Sinniah, A. (2019). Liking, sharing, commenting and reacting on Facebook: User behaviors’ impact on sentiment intensity. Telematics and Informatics, 39, 25–36. doi: 10.1016/j.tele.2018.12.005

Kesner, I. (1994). Executive succession: Past, present & future. Journal of Management, 20(2), 327–372. https://doi.org/10.1016/0149-2063(94)90019-1

Larsson, A. O. (2016). In it for the Long Run? Journalism Practice, 11(4), 438–457. https://doi.org/10.1080/17512786.2015.1121787

(32)

Li, X., Cox, A., & Wang, Z. (2018). How do social network sites support product users’ knowledge construction? A study of LinkedIn. Online Information Review, 42(3), 304– 323. https://doi.org/10.1108/oir-04-2017-0133

McDermott, B. Bill McDermott on LinkedIn: Choose your words carefully. The potential for words to change minds: 56 comments. LinkedIn.

https://www.linkedin.com/feed/update/urn:li:activity:6516331903002513408/.

Men, L. R., & Tsai, W.-H. S. (2016). Public engagement with CEOs on social media: Motivations and relational outcomes. Public Relations Review, 42(5), 932–942. https://doi.org/10.1016/j.pubrev.2016.08.001

Men, L. R., Tsai, W.-H. S., Chen, Z. F., & Ji, Y. G. (2018). Social presence and digital dialogic communication: engagement lessons from top social CEOs. Journal of Public Relations Research, 30(3), 83–99. https://doi.org/10.1080/1062726x.2018.1498341

Mutz, D. C. (2006). Hearing the other side: Deliberative vs. participatory democracy. New York: Cambridge University Press.

Oh, O., Rao, H. R. & Agrawal, M. (2013). Community Intelligence and Social Media Services: A Rumor Theoretic Analysis of Tweets During Social Crises, MIS Quarterly. 37, 407-426.

Ordenes, F. V., Grewal, D., Ludwig, S., Ruyter, K. D., Mahr, D., & Wetzels, M. (2018). Cutting through content clutter: How speech and image acts Ddive consumer sharing of social media brand messages. Journal of Consumer Research, 45(5), 988–1012.

(33)

Qi, J., Monod, E., Fang, B., & Deng, S. (2018). Theories of social media: Philosophical foundations. Engineering, 4(1), 94-102. doi:10.1016/j.eng.2018.02.009

Rooderkerk, R. P., & Pauwels, K. H. (2016). No comment?! The drivers of reactions to online posts in professional groups. Journal of Interactive Marketing, 35, 1–15. doi:

10.1016/j.intmar.2015.12.003

Saxton, G. D., & Waters, R. D. (2014). What do stakeholders like on Facebook? Examining public reactions to nonprofit organizations’ informational, promotional, and

community-building messages. Journal of Public Relations Research, 26(3), 280–299. doi: 10.1080/1062726x.2014.908721

Searle, J. R. 1969. Speech Act: An Essay in the Philosophy of Language, Cambridge: Cambridge University Press.

Searle, J. R. (1976). A classification of illocutionary acts. Language in Society, 5(1), 1–23. https://doi.org/10.1017/s0047404500006837

Simon, S., & Dejica-Cartis, D. (2015). Speech acts in written advertisements: Identification, classification and analysis. Procedia - Social and Behavioral Sciences, 192, 234–239. https://doi.org/10.1016/j.sbspro.2015.06.033

Stieglitz, S., & Dang-Xuan, L. (2013). Emotions and information diffusion in social media— sentiment of microblogs and sharing behavior. Journal of Management Information Systems, 29(4), 217–248. doi: 10.2753/mis0742-1222290408

Stiles, W. B. 1992. Describing Talk: A Taxonomy of Verbal Response Modes, Newbury Park, CA: Sage.

(34)

Schwartz, R., & Halegoua, G. R. (2014). The spatial self: Location-based identity performance on social media. New Media & Society, 17(10), 1643-1660.

doi:10.1177/1461444814531364

Sy, T., Côté, S., & Saavedra, R. (2005). The Contagious Leader: Impact of the Leader's Mood on the Mood of Group Members, Group Affective Tone, and Group Processes. Journal of Applied Psychology, 90(2), 295–305. https://doi.org/10.1037/0021-9010.90.2.295

Van Dijck, J. V. (2013). ‘You have one identity’: performing the self on Facebook and

(35)

Footnote

1 _{For further information on the umbrella project, please contact Jeroen Jonkman} (j.g.f.jonkman@uva.nl) and Ward van Zoonen (w.vanzoonen@uva.nl).

(36)

Appendix A

Intercoder Reliability Scores

Percent Agreement Krippendorff's Alpha (nominal) N Agreements N Disagreements N Cases N Decisions ICR 1 Assertive 70.9% 0.42 39 16 55 110 Directive 92.7% 0.81 51 4 55 110 Commissive 89.1% 0.44 49 6 55 110 Expressive 81.8% 0.56 45 10 55 110 Declarative 76.4% 0.33 42 13 55 110 ICR 2 Assertive 83.6% 0.58 46 9 55 110 Directive 100.0% 1.00 55 0 55 110 Commissive 92.7% 0.71 51 4 55 110 Expressive 87.3% 0.69 48 7 55 110 Declarative 76.4% 0.42 42 13 55 110 ICR 3: Assertive 90.00% 0.68 45 5 50 100 Directive 94.00% 0.87 47 3 50 100

(37)

Commissive 96.00% 0.73 48 2 50 100

Expressive 90.00% 0.76 45 5 50 100

(38)

Appendix B

Negative Binomial Regression Output for the Kudo Type Like

Parameter

Hypothesis Test

Exp(B)

95% Wald Confidence Interval for Exp(B) Wald

Chi-Square df Sig. Lower Upper

(Intercept) 1818.17 1 0.000 501.29 376.69 667.12 Assertive 1.88 1 0.171 0.84 0.66 1.08 Directive 5.52 1 0.019 0.74 0.57 0.95 Commissive 1.25 1 0.264 0.80 0.54 1.18 Expressive 0.46 1 0.499 0.91 0.70 1.19 Declarative 3.72 1 0.054 1.31 1.00 1.72 length 9.51 1 0.002 1.00 1.00 1.00 Followers 65.78 1 0.000 1.00 1.00 1.00

(39)

Appendix C

Negative Binomial Regression Output for the Kudo Type Celebrate

Parameter

Hypothesis Test

Exp(B)

(40)

Appendix D

Negative Binomial Regression Output for the Kudo Type Love

Parameter

Hypothesis Test

Exp(B)

(41)

Appendix E

Negative Binomial Regression Output for the Kudo Type Curious

Parameter

Hypothesis Test

Exp(B)

(42)

Appendix F

Negative Binomial Regression Output for the Kudo Type Insightful

Parameter

Hypothesis Test

Exp(B)

(43)

Appendix G

Negative Binomial Regression Output for the Kudo Type Like with Moderators Sentiment

Parameter

Hypothesis Test

Exp(B)

(Intercept) 922.00 1 0 499.82 334.66 746.48 Assertive 1.26 1 0.261 0.80 0.55 1.18 Directive 1.63 1 0.202 0.80 0.56 1.13 Commissive 0.06 1 0.813 1.08 0.58 2.00 Expressive 0.00 1 0.951 0.99 0.68 1.43 Declarative 2.84 1 0.092 1.42 0.94 2.14 Negative Sentiment 1.26 1 0.261 5.27 0.29 95.49 Positive Sentiment 0.03 1 0.87 0.95 0.51 1.76 Assertive*Negative Sentiment 0.41 1 0.524 1.72 0.32 9.16 Assertive* Positive Sentiment 0.23 1 0.632 1.13 0.69 1.87

(44)

Directive* Negative Sentiment 0.08 1 0.778 0.68 0.05 9.84 Directive* Positive Sentiment 0.58 1 0.447 0.81 0.47 1.40 Commissive* Negative Sentiment 4.31 1 0.038 0.05 0.00 0.85 Commissive* Positive Sentiment 1.46 1 0.226 0.61 0.27 1.36 Expressive* Negative Sentiment 2.03 1 0.155 0.15 0.01 2.07 Expressive* Positive Sentiment 0.03 1 0.863 0.95 0.54 1.69 Declarative* Negative Sentiment 1.44 1 0.23 0.29 0.04 2.17 Declarative* Positive Sentiment 0.11 1 0.736 0.91 0.54 1.55 length 8.09 1 0.004 1.00 1.00 1.00 Followers 62.50 1 0.000 1.00 1.00 1.00

(45)

Appendix H

Negative Binomial Regression Output for the Kudo Type Celebrate with Moderators Sentiment

Parameter

Hypothesis Test

Exp(B)

(Intercept) 17.68 1 0.000 4.70 2.28 9.65 Assertive 0.29 1 0.591 1.18 0.65 2.16 Directive 9.39 1 0.002 0.43 0.25 0.74 Commissive 0.81 1 0.367 0.65 0.26 1.65 Expressive 0.06 1 0.805 1.08 0.57 2.06 Declarative 7.31 1 0.007 2.76 1.32 5.77 Negative Sentiment 0.14 1 0.71 2.33 0.03 203.65 Positive Sentiment 0.58 1 0.445 0.68 0.25 1.83 Assertive*Negative Sentiment 0.36 1 0.548 0.42 0.03 7.03 Assertive* Positive Sentiment 2.04 1 0.153 1.77 0.81 3.90

(46)

(47)

Appendix I

Negative Binomial Regression Output for the Kudo Type Love with Moderators Sentiment

Parameter

Hypothesis Test

Exp(B)

(Intercept) 14.16 1 0.000 3.86 1.91 7.81 Assertive 3.97 1 0.046 0.54 0.29 0.99 Directive 32.91 1 0.000 0.18 0.10 0.32 Commissive 2.09 1 0.148 0.49 0.19 1.29 Expressive 0.28 1 0.597 0.84 0.44 1.61 Declarative 3.23 1 0.073 1.95 0.94 4.02 Negative Sentiment 0.06 1 0.801 1.83 0.02 197.98 Positive Sentiment 1.60 1 0.206 0.52 0.18 1.44 Assertive*Negative Sentiment 0.10 1 0.757 0.65 0.04 10.23 Assertive* Positive Sentiment 6.83 1 0.009 2.92 1.31 6.53

(48)

(49)

Appendix J

Negative Binomial Regression Output for the Kudo Type Curious with Moderators Sentiment Parameter Hypothesis Test Exp(B) 95% Wald Confidence Interval for Exp(B) Wald

(Intercept) 0.13 1 0.721 0.87 0.39 1.91 Assertive 0.02 1 0.891 0.95 0.44 2.03 Directive 3.41 1 0.065 0.49 0.23 1.04 Commissive 0.60 1 0.437 0.61 0.17 2.14 Expressive 0.08 1 0.774 0.90 0.42 1.90 Declarative 0.32 1 0.574 0.78 0.32 1.88 Negative Sentiment 0.00 1 1 ######## 0.00 .b Positive Sentiment 3.90 1 0.048 0.25 0.06 0.99 Assertive*Negative Sentiment 0.00 1 1 0.00 0.00 .b Assertive* Positive Sentiment 2.71 1 0.1 2.39 0.85 6.72 Directive* Negative Sentiment 0.00 1 1 0.00 0.00 .b

(50)

Directive* Positive Sentiment 0.40 1 0.529 1.49 0.43 5.12 Commissive* Negative Sentiment 0.00 1 1 0.00 0.00 .b Commissive* Positive Sentiment 0.83 1 0.364 2.20 0.40 12.03 Expressive* Negative Sentiment 0.00 1 1 0.00 0.00 .b Expressive* Positive Sentiment 0.08 1 0.778 1.20 0.34 4.28 Declarative* Negative Sentiment 0.00 0.00 0.00 Declarative* Positive Sentiment 0.78 1 0.378 1.69 0.53 5.40 length 0.02 1 0.877 1.00 1.00 1.00 Followers 31.66 1 0.000 1.00 1.00 1.00

(51)

Appendix K

Negative Binomial Regression Output for the Kudo Type Insightful with Moderators Sentiment

Parameter

Hypothesis Test

Exp(B)

95% Wald Confidence Interval for Exp(B)

Wald

(Intercept) 23.24 1 0 0.13 0.06 0.29 Assertive 5.08 1 0.024 2.54 1.13 5.70 Directive 0.04 1 0.848 1.08 0.49 2.37 Commissive 3.42 1 0.064 0.22 0.05 1.09 Expressive 5.29 1 0.021 2.46 1.14 5.29 Declarative 3.02 1 0.082 0.44 0.18 1.11 Negative Sentiment 0.00 1 1 0.00 0.00 .b Positive Sentiment 1.27 1 0.259 2.30 0.54 9.76 Assertive*Negative Sentiment 0.00 0.00 0.00

(52)

Assertive* Positive Sentiment 0.30 1 0.586 0.73 0.23 2.30 Directive* Negative Sentiment 0.00 0.00 0.00 Directive* Positive Sentiment 1.65 1 0.199 0.43 0.12 1.57 Commissive* Negative Sentiment 0.00 1 0.975 0.00 ######## ######## Commissive* Positive Sentiment 4.09 1 0.043 8.60 1.07 69.18 Expressive* Negative Sentiment 0.00 1 1 4345.27 0.00 .b Expressive* Positive Sentiment 5.01 1 0.025 0.22 0.06 0.83 Declarative* Negative Sentiment 0.00 0.00 0.00 Declarative* Positive Sentiment 0.70 1 0.405 1.71 0.48 6.09 length 9.00 1 0.003 1.00 1.00 1.00 Followers 56.90 1 0.000 1.00 1.00 1.00

(53)

Appendix L

Negative Binomial Regression Output for the Kudo Type Like with Moderator Thought Leadership (Intercept) 1637.00 1 0.000 454.88 338.18 611.86 Assertive 0.34 1 0.561 0.92 0.68 1.23 Directive 4.71 1 0.03 0.72 0.54 0.97 Commissive 0.21 1 0.645 0.89 0.53 1.48 Expressive 0.01 1 0.932 1.01 0.76 1.34 Declarative 6.42 1 0.011 1.50 1.10 2.06 Assertive*Thought Leadership 0.23 1 0.633 0.89 0.56 1.43 Directive*Thought Leadership 1.00 1 0.316 1.30 0.78 2.19 Commissive*Thought Leadership 0.08 1 0.779 0.89 0.41 1.95 Parameter Hypothesis Test Exp(B) 95% Wald Confidence Interval for Exp(B) Wald