Following science on social media: The effects of humor and source likability

Humor in science on social media

There are many types, or techniques, of humor. In advertising research, a content analysis of over 300 humorous television commercials identified seven categories into which humor types cluster (Buijzen and Valkenburg, 2004). These categories are slapstick, surprise, irony, clownish humor, satire, misunderstanding, and parody. Others, in synthesizing research, have identified 11 forms of humor: irony, satire, sarcasm, over- and understatement, self-deprecation, teasing, replies to rhetorical questions, clever replies to serious statements, double entendre, transformations of well-known sayings, and parody (Martin, 2010).

Funny science content is common on social media, including the hashtags #overlyhonestmethods, #fieldworkfail, and #reviewforscience, which have become particularly popular among those in the scientific community. Such hashtags are used to identify satirical posts about science, primarily related to scientific methods (for an empirical analysis of #overlyhonestmethods, see Simis-Wilkinson et al., 2018). We began this project by conducting a preliminary content analysis to empirically examine the types and relative volumes of humor present in science content on Instagram and Twitter. Among the categories of humor found, ¹ wordplay, satire, and anthropomorphic humor were most prevalent. In this study, we opted to focus on wordplay and anthropomorphism as these humor types are easily manipulated without altering other aspects of the surrounding content.

Anthropomorphic humor is the attribution of human characteristics to animals or objects that lack human qualities (Buijzen and Valkenburg, 2004). Wordplay, which consists primarily of puns, is the humorous use of words that evoke second meanings and involves playing with meanings of words or different words with similar pronunciations. Before examining the effect of humor types on our downstream variables, we first establish that a science message designed to be funny, relative to a non-humorous one, causes viewers to perceive or experience humor.

Humor has many conceptualizations and operationalizations. Often, survey respondents are asked to indicate the extent to which they perceive humor in a stimulus they have viewed (Eisend, 2009; Nabi, 2016). In other words, humor is often conceptualized and operationalized as how much mirth or amusement people experience when viewing content. While it has also been measured by categorizing stimuli by the presence or absence of humorous features (Duncan and Nelson, 1985), mirth is a more appropriate operationalization of the psychological response to humor for this study as it embodies an affective response to multiple humor types (Gulas and Weinberger, 2006). Unlike laughter, mirth does not restrict the definition of humor response to physiological reactions. Thus, we conceptualize and operationalize humor as a psychological, instead of a physiological, response, and measure the extent to which participants perceive the stimulus to be funny.

Factual knowledge as a moderator

Humor theorists generally propose three cognitive processes related to the recognition, appreciation, and comprehension of humor, or “getting the joke” (Speck, 1991; Suls, 1983). These are (a) incongruity resolution, (b) tension relief or arousal safety, and (c) humorous disparagement. Although some scholars distinguish between joke recognition and appreciation (i.e. one might recognize a joke but not appreciate it, or fail to find it funny; Ritchie, 2018), we do not distinguish between recognition and appreciation or comprehension. In this work, we focus on incongruity resolution as a theoretical underpinning as it is the framework best supported by empirical evidence (Uekermann et al., 2007). ² The incongruity resolution model was first proposed by Suls (1972) and is a problem-solving approach to processing funny information. In the first stage of this two-stage model, a viewer encounters an incongruity, usually between the punchline of the joke and their expectations about the information. In the second stage, the information consumer engages in a form of problem-solving in which they reconcile the incongruity. In this model, the content of the funny message activates a body of contextual knowledge used to interpret the information (Suls, 1972, 1983).

Several theoretical frameworks for understanding humor elicitation and appreciation also identify knowledge as integral to the process (Wyer and Collins, 1992). A recent framework, the encryption model of humor (Flamson and Barrett, 2008), is based on relevance theory (Sperber and Wilson, 1995) and proposes that knowledge is the encryption key that unlocks and allows people to make sense of jokes, that is, the encrypted message. Humor in this model represents a signal of common knowledge, values, and attitudes. In this sense, knowledge is key to appreciating and producing humor. In other words, the extent to which a viewer finds a joke to be funny depends on whether they possess the encryption key, or knowledge, to “get the joke.” Importantly, proponents of the encryption model highlight that it is not at odds with the incongruity resolution model. Instead, incongruity is one means of encryption and signaling shared knowledge among humor producers and receivers.

Empirical evidence supports that knowledge is an integral contextual factor in joke appreciation. In a case study of young people’s responses to cartoons about the issue of gay marriage, El Refaie (2011) found that enjoyment of the joke was partly dependent on background knowledge of the consumers. Evidence for knowledge as a moderator of humor appreciation can also be found in studies of political satire. In a Dutch study examining people’s attitudes toward budget cuts on public broadcasting, the availability of background knowledge was found to moderate the effect of satire on perceived humor and people’s political preferences (Boukes et al., 2015). Given extant evidence, we hypothesize that the extent to which participants find our stimulus materials to be funny will depend on their levels of science knowledge.

Humor, perceived source likability, and persuasion

Research on the effects of humor and mirth on behavioral intentions have been primarily conducted in the context of entertainment (e.g. Futerfas and Nan, 2017), health communication (e.g. Nabi, 2016), advertising (e.g. Yoon, 2015), and environmental communication (Skurka et al., 2019). In entertainment media, researchers exposed female viewers to one of three conditions—a humorous entertainment narrative about unprotected sex, a version of the narrative without humor, or a control condition unrelated to unprotected sex (Futerfas and Nan, 2017). The humorous narrative was found to reduce intentions to engage in unprotected sex. A meta-analysis of humor in advertising found collective evidence for humor’s impact on behavioral intentions, specifically purchase intentions (Eisend, 2009). Recent scholarship that supports this conclusion has examined the effect of humor in online video advertisements (Goodrich et al., 2015) and in Snapchat (a multimedia social media app) ads (Phua & Kim, 2018). In an examination of McDonald’s ads, researchers found that intentions to share the ad with others were positively impacted by humor (Sabri and Michel, 2014). In health communication, humor has been found to indirectly influence cancer self-examination behavioral intentions by reducing anxiety and improving attitudes toward self-exams (Nabi, 2016). In addition, recent work on climate change shows that humor can increase intentions to engage in climate activism relative to an off-topic control message (Skurka et al., 2019).

This study focuses on motivation to follow more science on social media as an outcome of the persuasive impact of a funny message. The participatory nature of social media allows its users to actively engage with experts to access new ideas in specific fields. Twitter enables individuals to stay abreast of scientific and technological advances by receiving updates from whatever set of subject experts they choose to “follow.” Features of content have been shown to motivate individuals to follow social media accounts. For example, Perrault et al. (2019) identified five factors that motivate students to follow their college’s health center accounts, including incentives, relevance, knowledge of the existence of the account, message aesthetics, and popularity. Pertinent to the present work, respondents mentioned that “funny posts” and “fun articles” posted on the social media account serve as motivations for following it. In another study using Instagram, reasons such as “it is fun” and “entertaining” were identified as motivators for following political leaders on Instagram (Parmelee and Roman, 2019).

There is also evidence that the humorousness of a scientific message predicts people’s intentions to engage with such content (Yeo et al., 2020). Given the existing evidence, we hypothesize the following:

Humor can impact how audiences make sense of a message as well as things connected to that content (e.g. message source; Gulas and Weinberger, 2006). In the context of advertising, humor has been found to result in both lower and higher order outcomes (Duncan and Nelson, 1985). Lower order outcomes are concerned with audiences’ immediate evaluation of a message, while higher order outcomes include actions or cognitions activated in response to viewing humorous material. Substantial evidence points to a relationship between humor and lower order outcomes, such as attitudes toward an advertisement, attention paid to an ad, positive affect, and ad recall (Duncan and Nelson, 1985; Eisend, 2009; Sabri, 2012; Weinberger et al., 1995).

Of specific interest to this study is research showing humor influences one’s evaluation of the source of that humor (Markiewicz, 1974; Sternthal and Craig, 1973). The relationship between humor and source likability is relatively consistent. Humor has been linked to more favorable teacher evaluations (Bryant et al., 1980), increased liking of educational materials (Bryant et al., 1981) and educational television programs (Zillmann and Bryant, 1980), increased liking of ads (Duncan and Nelson, 1985), and higher favorability scores for brands (Duncan and Nelson, 1985). In particular, inoffensive humor has been associated with attraction and the building of rapport between individuals (Wilson, 1979). When someone makes another person laugh, the recipient associates the source of humor with the pleasure of laughing. As a result, they view the source as more likable (Graham et al., 1992). Humorous people are generally rated more favorably than others, a finding that has been replicated across diverse contexts (Wanzer et al., 1996), including in experimental approaches where speeches were designed to vary based on the presence or absence of jokes (Gruner, 1967, 1970). Given the findings noted earlier, we propose the following:

Audience perceptions of the communicator have long been recognized as an important influence on the effectiveness of communication (Hovland and Weiss, 1951). Among the various desirable attributes of message sources, communicator likability plays an arguably decisive role in the persuasive impact of a message. Often conceptualized as an affective evaluation linked to an object (Roskos-Ewoldsen and Fazio, 1992), perceived source likability is associated with traits that make a person likable in a general sense, but that are not necessarily relevant to the person’s expertise or status as an authority on the topic at hand (Stone and Eswara, 1969). Researchers have consistently found that communicators who were perceived as attractive and likable were more likely to influence audiences’ views through explicitly expressed intentions to persuade (Mills and Aronson, 1965; Reinhard et al., 2006).

Roskos-Ewoldsen and Fazio (1992) argued that the persuasive effect of source likability is moderated by its accessibility from memory. In other words, the more accessible the source’s likability is, the greater its impact on attitudinal changes. They propose mechanisms that might underpin the persuasive impact of source likability, once it is activated from memory (see also Roskos-Ewoldsen et al., 2002), based on the heuristic-systematic (HSM; Chaiken, 1980) and elaboration likelihood models (ELM; Petty and Cacioppo, 1986). The likable traits of the communicator may act as heuristic cues for attitudinal judgments (Basnyat and Lim, 2018). For instance, Basnyat and Lim found that young women in Singapore relied on source likability as a heuristic cue when deciding whether to seek the human papillomavirus (HPV) vaccination. When a source’s likability is more accessible from memory, heuristic processing becomes more likely, which in turn has an impact on whether people will agree with the message (see Roskos-Ewoldsen and Fazio, 1992; Roskos-Ewoldsen et al., 2002). In fact, Chaiken and Eagly (1983) found that, in both video and audio conditions, persuasive messages from likable sources were more likely to motivate attitudinal changes because personal attributes were more salient for heuristic processing. Interestingly, the persuasive effect of messages from unlikable sources was strongest in the written communication condition, possibly because the relatively weak source cues in that condition led to systematic processing instead.

Most empirical studies have noted a positive link between the perceived likability of the communicator and the persuasive impact of the communication messages (e.g. Basnyat and Lim, 2018; Mills and Aronson, 1965; Reinhard and Messner, 2009). Importantly, persuasive impact is not limited to attitudinal judgments such as product and company evaluations, but can be translated into behavioral intentions such as purchase intentions (Reinhard and Messner, 2009). In the context of this study, we propose that

Taken together, our hypotheses and research question link our antecedent and consequent variables in a path model that is moderated by factual knowledge. Thus, we test a conditional effects model:

Experimental design and measures

The main experiment employed a 4 (humor type) × 2 (social media metrics) between-subjects design. Social media metrics were manipulated by changing the number of retweets and likes associated with the original tweet. In the low social media metrics condition, the original post had three retweets and five likes; the post in the high social media metrics condition had 288 retweets and 480 likes. A secondary experiment concerning the measurement of factual science knowledge consisted of two conditions; one in which knowledge was measured on a 3-point scale (“True,” “False,” and “Don’t Know”) and another that employed a 5-point scale ranging from “Definitely true” to “Definitely false.” In the analysis presented here, we controlled for the social media metrics (0 = “low metrics,” 1 = “high metrics”) and knowledge scale (0 = “3-point scale,” 1 = “5-point scale”) manipulations by including two dummy variables. ⁴ For both the primary and secondary experiments, respondents were randomly assigned to experimental conditions.

Although we report the effects of the humor manipulation on respondents’ levels of experienced humor, or mirth, elsewhere (Yeo et al., 2020), we briefly repeat these results here for convenience. The level of humor induced by a message is subjective. This results in an important methodological choice; while we manipulate the type of humor in the experimental stimuli, using the humor type conditions as the independent variable would not necessarily account for differences in the amount of humor that respondents might perceive (Duncan and Nelson, 1985). Indeed, individuals might respond to the same funny message in different ways. Therefore, we use mirth a mediating variable as it is the psychological state experienced by respondents as a result of the experimental conditions and we are interested in the downstream effects of this psychological state (O’Keefe, 2003).

Participants were randomly exposed to one of eight screenshots of a conversation on Twitter about science. Both an image and text were included to reinforce the humor manipulation. Each condition contained one response to the original post by a fictional user, Kasey Chase, who was consistent with the humor type condition. In all conditions, the Twitter conversation was started by Dr Jamie Devon, a fictional scientist. All names in the manipulation were gender ambiguous.

The four humor conditions were no humor, anthropomorphism, wordplay, and combined, which was a combination of anthropomorphic humor and wordplay. In the no humor condition, the text of the post was altered to contain a science fact. We strived to keep the number of words in the post consistent (18–20 words not including hashtags) between conditions. Anthropomorphism was manipulated in the image of the post; it included a mouth, eyes, and arms on the atoms. The wordplay manipulation occurred in the text of the post. The combined humor condition consisted of both text and image manipulations. Experimental conditions can be found in the Supplemental Material. ⁵

After exposure to the stimulus, respondents were asked questions designed to assess humor evoked by the Twitter conversation as well as perceived popularity of the tweet and motivation to follow more science on social media. All variables used in this study were measured after exposure to the stimulus.

Mirth was operationalized by asking respondents to indicate how they would describe the Twitter conversation on 5-, 7-point semantic differential scales (not humorous-humorous, not funny-funny, not playful-playful, not amusing-amusing, not entertaining-entertaining). We averaged these items to create an index (Cronbach’s α = .937, M = 4.49, SD = 1.89).

Perceived source likability was an averaged index of four items (Cronbach’s α = .92, M = 4.85, SD = 1.43) asking respondents to consider the scientist who started the Twitter conversation and rate their agreement with the following statements on a 7-point Likert-type scale (1 = “Strongly disagree,” 7 = “Strongly agree”): (1) This person seems friendly, (2) This person seems likable, (3) This person seems warm, (4) This person seems approachable.

We operationalized factual knowledge by asking respondents seven questions about general science knowledge. The seven knowledge questions were selected from those used in the National Science Board’s (2018) Science and Engineering Indicators: “The center of the Earth is very hot” (true), “The continents have been moving their location for millions of years and will continue to move” (true), “All radioactivity is man-made” (false), “Electrons are smaller than atoms” (true), “Lasers work by focusing sound waves” (false), “It is the father’s gene that decides whether the baby is a boy or a girl” (true), and “Antibiotics kill viruses as well as bacteria” (false). Possible response categories were “Definitely true,” “Likely true,” “Likely false,” “Definitely false,” and “Don’t know.” Correct responses, regardless of whether the respondent chose the “Definitely” or Likely” option, were summed to create a 7-point index of factual knowledge (M = 3.85, SD = 1.78). Motivation to follow more science on social media was measured by asking respondents the extent to which the Twitter conversation (1) motivated them to pay more attention to science in social media, (2) motivated them to follow more scientists on social media, and (3) motivated them to pay closer attention to updates from scientists on social media (1 = “Not at all,” 7 = “Very much”; Cronbach’s α = .97, M = 3.58, SD = 2.04).

Data analysis

All analyses were conducted using IBM SPSS Statistics 26. We addressed our research questions and hypotheses using a conditional effects model with two serial mediators (Model 83; Figure 1) with the PROCESS 3.3 add-on for SPSS (Hayes and Matthes, 2009). Since our independent variable is nominal and multicategorical, we included the “mcx” option with indicator coding in our model (for more information, see the PROCESS 3.3 documentation).

OPEN IN VIEWER

Figure 1.

Conceptual model (PROCESS Model 83) with multicategorical independent variable, humor type.

To probe significant interactions in the model, we used floodlight analysis (Spiller et al., 2013), which uses the Johnson-Neyman (JN) technique (Johnson and Neyman, 1936). The SPSS add-on, PROCESS 3.3, is unable to conduct a floodlight analysis when the independent variable is multicategorical because deriving regions of significance for an interaction becomes highly complex (for a detailed explanation, see Hayes and Montoya, 2017). In this analysis, we are interested in the interactions between a specific humor condition and factual knowledge influencing mirth rather than an omnibus floodlight analysis. Therefore, we follow the procedure outlined in Hayes and Montoya (2017) for conducting pairwise inferences using the JN method, and PROCESS Model 83 with 10,000 bootstrap samples.