Over the span of three weeks, we attempted to study the impact that various social media
platforms had on the production of science communication. We had already surveyed the
literature and found convincing evidence that science communication could play a
significant role in the public education of science. For example, Wilson et al. (2017)
had already highlighted how reliable, accurate science communication could improve public trust in expert opinion.
Our specific research question was whether the choice of a social media platform had significant
influence over how science communication posts were conducted, received, and shared. There
was some prior evidence to suggest such an effect. Warden (2010) discussed the unique
advantage of blog posts, as they are more in-depth than mainstream science journalism
articles but still shorter and more accessible than journal publications. This
suggested that platforms with a restriction on word count (e.g., Twitter) would have
to be used differently than blog platforms (e.g., Medium, Tumblr). Further support came
from Hwong et al. (2017), who had studied the use of Twitter and
Facebook and found that visual elements were a strong driver of engagement. We
hypothesized that the aesthetic appeal of our different visuals across all platforms
would be a useful property for comparison.
We noted that the blogging platforms Medium and Tumblr were most suited for consistent
engagement. Posts on these platforms continued to get attention days after they were
initially posted, which was significantly different from rival platforms. This effect
was particularly noticeable following the first post, when continuous views later in
the week significantly exceeded posts on rival platforms.
We also noted that the second post underperformed on Medium relative to both other posts on the
platform and the same post on Tumblr. This may have occurred due to a demographic difference
between the platforms. According to the US statistical research firm Statista, up to a fifth of
Tumblr users are millennials less than 35 years of age, who may be more inclined to engage with
science communication that uses a ‘cool factor’ to hook the audience (Statista). This hypothesis
is supported by research from the digital marketing agency MADX (MADX, 2021).
Meanwhile, posts on Twitter and TikTok showed the highest rate of engagement in the immediate 24–48
hours after posting, followed by a plateauing of engagement metrics. The second Twitter post was
an exception to this rule: after it was retweeted by an account with a lot of followers, it
quickly dominated that week’s rankings. This demonstrates the unpredictability and time-sensitivity
of the Twitter and TikTok algorithms relative to the consistently growing engagement of Tumblr and Medium.
As expected, there was a tradeoff between the appealing visual elements and the level of detail
in the exposition. Whereas Medium and Tumblr allowed for longer, detailed explanations of physics,
Twitter, Instagram, and TikTok required us to stick to key words and visually appealing images to
deliver the greatest ‘bang for buck.’ The use of key words was a strategy from Pavlov et al. (2018)
and the emphasis on visuals was informed by experimental results from Hwong et al. (2017).
On TikTok, we observed that narration quickly ate up the short three-minute time limit allotted
to videos at the time. However, we also observed the greatest potential for reaching a wide
audience. Both the second and third posts received the most engagement on TikTok, and e
ngagement on the platform as a whole grew by a factor of 100 from six views on the first
post to 1622 views on the third. We suggest this growth was driven by an increasingly
simple video production style emphasizing narration over embedded video clips and
demonstrations, as well as catchier thumbnail visuals. This suggests that visuals are
more important than content for engagement on TikTok, which parallels what
Hwong et al. (2017) theorize about Twitter.
On Twitter, we were initially even more constrained to the 280-character limit.
We were able to circumvent this somewhat by placing a substantial amount of text
in embedded graphics and videos, as well as using Twitter’s thread feature to string
together multiple tweets, although our analysis showed that only up to half of the users
who engaged with the first tweet read through the entire thread.
On Instagram, there was much freedom with the structure of the posts, having options for an
in-feed post of up to 10 slides, a story post which lasts for 24 hours, a reel (a
short video, up to 60 seconds), or even an IG TV (a longer video, up to 60 minutes).
We selected the in-feed option since it is the most commonly used and longest standing
feature of Instagram, including image and text heavy content, and combination
of the text and images for our three posts. We noticed a downward trend in the
engagement on the three posts, with 53 total likes on the first post, and 18 on
the second and third. Delving into the second two posts, there was an increase
in the algorithm engagement on the third post, mostly from the accompanying hashtags.
In reviewing this, it must be noted that the hashtag dependency of the in-feed posts
is very flawed, with an average of only 20% of the hashtags on posts being relevant
to the actual content (Giannoulakis & Tsapatsoulis, 2019), meaning this is not an
efficient way of sharing the information for the hopes of communicating science.
Beyond this, the Instagram reels have their own algorithm (Mosseri, 2021)
and would have provided a unique opportunity to investigate engagement in that regard
as well if we followed that route.
When comparing between platforms, we also noted that it was difficult to gauge which post
had performed the best overall. Our analysis varied depending on the engagement metric
selected. For instance, the second post received significantly more ‘likes’ on Tumblr
than rival platforms, but the same post was viewed 7.9 times more on TikTok, even
though the Tumblr post was the most-viewed among the others. We also noted that our
posts did not receive many comments on any platform despite the evidence of their
popularity, suggesting that comments may not be a reliable metric of engagement.
We conclude that engagement should not be understood as a single metric but as a family
of metrics that each imply something different about a post. This is consistent with
the literature, where marketing researchers have outlined various metrics that yield
different conclusions (Trunfino and Rossi, 2021). For instance, the number of views
indicates how successfully a post was picked up by the platform’s algorithm, but the
number of likes is a stronger indicator of whether it was interesting enough to merit
significant attention. Attention can also be understood on a more fine-grained level.
The number of times a post is shared indicates a desire to spread the content, and
apps may also offer analytics for how much of a post user’s actually read, such as
scroll depth on a blog post or how many tweets in a thread are opened.
In conclusion, we found that science communication could be conducted successfully on
a number of different social media platforms. Each platform has its own strengths
and weaknesses, chief among which is the tradeoff between the level of detail offered
by blogging platforms like Medium and Tumblr and the ability to reach a wider audience
with engaging visuals on Twitter, Instagram, and TikTok. Among these later platforms,
our most interesting results included the observation that simplistic videos with
appealing thumbnail visuals can attract the greatest audience on TikTok; the data
showing that at most half of the users who engage with a Twitter thread will read to
the end; and more image heavy posts with little text will yield more interaction than
text heavy posts. Engagement on each of these platforms can be measured with different
metrics which all indicate different user experiences. Therefore, while it is possible
to engage the public in science communication on any platform, a successful science
communication strategy will require choosing the platform and engagement metrics
that best suit one’s goals.