Likes, comments, and shares of marine organism imagery on Facebook

Likes

For likes, most activity occurred within the first 24 h after posting; 75% of image posts received 81.3% of their likes within the first day (Fig. 2). Likes within this 24-hour time span ranged from 12 to 727 (Fig. 2).

An analysis of the Facebook likes indicates the data are overdispersed with respect to a Poisson model (p ≪ 0.0001). A negative binomial regression was thus utilized to analyze the data. Analysis indicates that only taxa (p ≪ 0.0001), image type (p = 0.0066), and awe factor (p ≪ 0.0001) were significant predictors of Facebook likes on images. A full model with both significant and non-significant predictors (AIC = 725) yields a pseudo-R² of 0.71, while the reduced model (AIC = 720) with only significant factors yielding a pseudo-R² of 0.70.

Images of Cephalopods and Osteichthyes generated significantly more likes that many other taxonomic groups including Echinoderms, Nematodes, Pinnipeds, and Cetaceans (p-value: ≫0.0001–0.0093, Fig. 3, Appendix S1). Osteichthyes also received significantly more likes than Polychaetes, Nudibranchs, Poriferans, Anthozoans, and Testudines (p-values: 0.0003–0.0093, Fig. 3, Appendix S1). Medusozoans also received more likes than Pinnipeds (p = 0.0090, Fig. 3, Appendix S1). On average, image posts of Medusozoans received 69.6, Cephalopods 115.5, and Osteichthyes 165.3 likes, whereas other taxa received between 22.3 and 56.6 likes.

Standard scientific images received significantly more likes than behavioral images (p = 0.0047, 55.2 versus 35.1 likes, Fig. 3, Appendix S1). Color images were not significantly different from the other two categories. Likewise, high awe images received more likes than low awe images (p ≫ 0.0001, 65.3 versus 33.1).

Using the predictive model for Facebook (see Appendix S1) likes and selecting for high-performing categories in taxa, awe, and photo type, an additional image was posted of a Cephalopod, in color, and with a high awe factor. The model predicted likes were 192 (range 93–290). The posted image of a bobtail squid on October 31, 2018 yielded 231 likes within 24 h after posting.

Shares

For shares, the range spanned from 0, incorporating 21% of posts, to 734 (Fig. 2). Unlike Facebook likes, however, only 25% of posts received 80.8% of their shares within the first 24 h after posting (Fig. 2).

The data for Facebook shares on image posts was heavily zero-inflated and both Poisson (p = 0.0058) and negative binomial regression models (p ≪ 0.0001) were overdispersed. A zero-inflation negative binomial regression model was utilized to account for the distribution of the count data. Because zero-inflation models run two models on the zero and non-zero data, they are particularly sensitive to the number of states in independent variables and the number of variables. Taxa were culled into post hoc reduced taxonomic variable based on Facebook like performance into a low, average, and high performance category based on groups identified in pairwise comparisons. Cephalopods, Osteichthyes, and Medusozoans were placed in the high category; Gastropods/Bivalves, Elasmobranchs, Cetaceans, Echinoderms, and Pinnipeds in the low category; and the rest of the taxa placed in the average category. In addition, independent variables were not included in the share model that were nonsignificant in the Facebook like model. A Facebook share model was constructed with photo type, awe factor, caption type, and the new three-category reduced taxonomic variable. Only awe factor (p = 0.0013), caption type (p = 0.0397), and the reduced taxonomic variable (p ≫ 0.0001) were significant (Fig. 4, Appendix S1). This model produced a pseudo-R² of 0.59.

Low, average, and high performance taxa received significantly different shares from each other (p = 0.0139-0.0423, means: 2.9, 8.2, 63.6). Although caption type was a significant factor, a Tukey post-hoc test did not yield a significant difference between shares of posts with public and scientific captions. High awe was significantly different in terms of shares from low awe images (p = 0.0137, 37.8 versus 12.0 average shares).

Comments

For comments, counts ranged from 0-52, but 36.8% of posts received no comments (Fig. 2). Of the remaining posts that received comments, a majority (67.2%) received those comments in the first 24 h after posting (Fig. 2).

Facebook comment data was also heavily zero-inflated, and Poisson (p = 0.0053) and negative binomial regression models (p ≪ 0.0001) were both overdispersed. For similar reasons as the share model, the comment model was reduced to include only photo type, awe factor, caption type, and the reduced taxonomic variable, producing a pseudo-R² of 0.32. Photo type (p = 0.0091), awe factor (0.0003), and the reduced taxonomic variable (0.0435) were all significant factors predicting comments but not caption type (p = 0.4150, Fig. 5, Appendix S1). Tukey post-hoc comparisons yielded no significant differences between different taxonomic categories or image types in terms of comments. High awe images received significantly more comments on average (3.68 versus 1.57, p = 0.0290).

Variability in interactions

In general, the Facebook audience in the study exhibited a wide range of engagement rates with the individual posts ranging from 12 to 727 likes, 0–734 shares, and 0–52 comments. Most of this activity, especially for likes and comments, occurred within the first 24 h after posting. For all three metrics, correlations between counts at 24 h after posting versus at 1000+ days were positive, significant, and high.

Comments

As expected, comment activity was much lower than likes and shares. In addition, many comments on the Facebook posts in this study were not commenting on the images but rather tagging another Facebook user. This pattern of liking and sharing versus commenting likely reflects the ease, a single button click, to either like or share on Facebook. The ease of superficially engaging with content has been coined “Slacktivism” and has been tied to the larger issue of “people who support a cause by performing simple measures (and) are not truly engaged or devoted to making a change” (Thaler et al., 2012). This pattern of low commenting is seen across DSN platforms; this study employs DSN’s Facebook page and the results are consistent with previous findings for the DSN blog. Previous research (Thaler et al., 2012) found that a majority of blog readers never commented (54.2%) and those that did comment left one or two comments (23.4%) over all posts they read. Surveyed readers of the blog, when asked why they did not comment, responded that they (1) did not feel qualified (28.6%), (2) had nothing to add (25.7%), or (3) did not generally comment on blogs (17.1%). Indeed, research indicates that commenting may be more heavily influenced by personality traits than social media content per se (Kabadayi & Price, 2014). Prior research also indicates that users’ active participation in social media, including commenting, requires practitioners to engage with users’ conversations at both the individual post and long-term levels (Camarero, Garrido & San Jose, 2018). The posts in this study were strongly visual, which may not have provided sufficient initial conversational stimulus for many users. Captions were also not specifically designed to prompt commenting, e.g., asking Facebook users a question. However, a previous study on the Facebook page of the Monterey Bay Aquarium Research Institute concluded that “an organization’s Facebook page does not seem to be a suitable space to trigger discussions” (Fauville et al., 2015).

Shares

Share behavior was more dynamic and unpredictable than either likes or comments. The sharing of posts occurred over longer lifespans, i.e., time decay of sharing is much longer than commenting or liking. Shares were also less predictable according to the criteria measured in this study. Pseudo-R² for the best share models was 0.32 versus 0.59 and 0.70 for comments and likes respectively. Sharing may reflect a robust social network component not captured in this study. High share counts require uptake by users with an extensive Facebook network, increasing the probability of share by their followers. The sharing network is affected by the size and structure of the Facebook user and their followers as well as positions of early sharers in the Facebook network (Weng, Menczer & Ahn, 2014). This user network effect is combined with Facebook’s content feed algorithm that weighs active interactions like commenting and sharing over passive interactions such as likes and click-throughs. Higher weighting is also assigned if users comment, share, and like the original share. To restate, shares require high activation energy but once a critical mass of shares and engagement on those shares is reached, then shared content appears more frequently in users’ feeds. This causes a positive feedback allowing other users to continue to share. Because of this activation energy required, lag times and volatility in shares may be greater. This is compounded by the fact that predicting what Facebook users share on the platform may also require knowledge of their self-interests and the perceived community incentives in their networks (Fu, Wu & Cho, 2017).

The role of charismatic megafauna

Traditional models of “charismatic megafauna” or “flagship species” dominate conservation and science communication efforts:

The focus on species in conservation has largely centered on vertebrates especially birds and large mammals. They are visible, dominant parts of our natural environments, and, for better or worse, extract more sympathy from the public than do most plants or insects. These “flagship” or “charismatic” species draw financial support more easily than do stinging insects or obscure mussels, and by so doing serve to protect habitat and other species under the “umbrella” of their large habitat requirements (Meffe & Carroll, 1997).

In the oceans, “charismatic megafauna” generally and largely incorporates marine mammals, sharks, and to a lesser degree fish and birds (Barney, Mintzes & Yen, 2005; Gerber & Heppell, 2004; Leader-Williams & Dublin, 2000). Indeed, in a survey of what taxa were considered charismatic, sharks, dolphins, and whales were included in the top ten (Albert, Luque & Courchamp, 2018). The use of “charismatic megafauna” has been criticized, including avocation of broadening the taxa promoted in conservation and communication (Entwistle, 2000; Guerra et al., 2011). The results of this study support these criticisms by suggesting that traditional definitions of “charismatic megafauna” and what organisms the public most engage with may be incorrect. Traditional outreach knowledge suggests that “charismatic megafauna” such as pinnipeds, cetaceans, and elasmobranchs may offer increased appeal to the public. However, across likes, comments, and shares, cephalopods, osteichthyes, and medusozoans out-performed these traditionally “charismatic megafauna”. These findings suggest the concept of “charismatic megafauna” may be different from and broader than previously thought.

Further support that the concept of “charismatic megafauna” may need reexamination is that individual images of taxa often perceived as not having public appeal performed well, including images of: two gastropods and one bivalve (63, 69, 73 likes), two polychaetes (65, 125), a nematode (77), and an arthropod (117). In addition, a monthly series focused on marine gastropods that occurred in 2015 on DSN’s Facebook page on average generated 50+ likes and 20+ shares per post, often higher than the more traditionally defined “charismatic megafauna” in this study. While the ongoing gastropod series may have raised awareness of post and inflated interactions, the posts do indicate that certain methods may overcome “charismatic megafauna” bias. These results and data also suggest that while some taxa may exhibit on average lower public engagement rates online, engagement can vary considerably. In other words, key and strategic posts on “non-charismatic taxa” still have the potential for high online engagement.

Quality and aesthetics of imagery

High-awe images consistently outperformed low-awe images by 1.6–3.2 times in likes, comments, and shares on Facebook. This suggests careful selection and curation of imagery is vital for online science engagement. Indeed, aesthetics contribute significantly to online engagement with image-based posts, including aspects of color, spatial arrangement, and complexity (Schifanella, Redi & Aiello, 2015). These results are also consistent with other research demonstrating that New York Times articles that connect emotionally with readers are more likely to be engaged with through comments, shares, and likes (Berger & Milkman, 2012). Specifically, content that evokes high physiological arousal, i.e., awe; a positive rather than a negative reaction; or an emotive reaction generates more engagement (Berger & Milkman, 2012; Kim, 2015; Nelson-Field, Riebe & Newstead, 2013). Awe while encompassing aspects of emotion may also incorporate novelty, and the combination of these are known to bolster content virality on Twitter (Vosoughi, Roy & Aral, 2018). For example, standard scientific images, with increased likes, often exhibited unknown animal features to Facebook users, i.e., novelty through informational discovery. The better performance of high-awe images in this study may benefit from both increased positive and emotive reaction as well as novelty. An image post of Mola mola, or sunfish, internal anatomy detailing the unique musculature and skeleton of the animal received the most number of likes and shares of all images deployed in the experiment: 727 and 734, respectively, in the first 24 h after posting.

Audience biases

The established audience of the DSN’s Facebook page, where this experiment occurred, could have impacted the generality of the results. Gender and age group are known to affect online engagement with specific content (Han et al., 2018). Privacy standards of Facebook prevent knowing the demographic composition of DSN’s Facebook followers. Although the overlap and similarities between the audience of DSN’s blog and Facebook page are unclear, biases in the demographics of the blog readership may also exist for the Facebook page. A prior survey of the DSN blog may provide some insights (Thaler et al., 2012). The typical DSN blog reader is a American young adult (25–34 years old; 41.5%) or middle-aged individual (35–54 years old; 34.5%). Readership was slightly more male (51.5%) and largely from the United States and Canada, followed by the United Kingdom (70.3%, 9.0%, and 4.0% respectively). The DSN blogs has a well-educated readership with a high percentage of readers having or pursuing a graduate or professional degree (17%). If these statistics hold for DSN’s Facebook audience as well then certain biases may affect the results. For example, the overall education of the audience and insular effect of attracting an audience with interests in the oceans may result in increased interest in a variety of taxa, ameliorating the “charismatic megafauna” effect.