Introduction & Inspiration

Many large institutions divide their content on YouTube across multiple channels depending on interests, subject, or even video tone, and the Smithsonian is no exception. In this report, I will examine the engagement data in aggregate and comparatively to see what the most and least popular content is among the channels, their engagement, and video posting rates. I explored this data set in another class previously, but I wanted to use this opportunity to refine my analysis and improve the visualizations.

My inspiration was this chart of data breaches. I felt that this chart does a good job of putting into perspective both the relative scale of various breaches, and just how many breaches there are in a way the raw numbers simply doesn’t convey to the human mind, as well as how they’ve been increasing (or at least, public knowledge of the breaches) over time.

Data

A sample of channels owned by the Smithsonian was selected by typing “Smithsonian” into the YouTube search box. The dataset was collected from YouTube using the YouTube Data Tools. As each channel was collected separately, the data was aggregated in Excel and visualized in Tableau.

Visualizations & Findings

This first chart is a distribution of postings by year, with bars to scale. This tells us some interesting things that scrolling though the raw Excel sheets or the video uploads won’t, such as how the Smithsonian Channel, despite being the main channel with by far the majority of videos uploaded, was only rarely utilized until 2012, implying a conscious change of direction at that point, as well as the abandonment of the Smithsonian Magazine channel in 2019. The Smithsonian Art Museum, however, has had a fairly consistent output despite being the second-least popular of the channels (as discussed below.)

These charts show and compare the total video views between the channels and their ratios of Likes to Dislikes. The Smithsonian Channel so engulfs the others in total views that it forms a razor-sharp mountain peak in this visualization. However, looking at the ratio of Likes to Dislikes, we can see that it is middle-of-the-pack in terms of reception among those that do watch the content- and interestingly, the Art Museum’s audience, though small, are big fans of its content.

The charts in this gallery show the ratio of Comments to Views, which is an indicator of how invested viewers are in a topic. Note that this cannot determine the tone of that engagement positive or negative, only the presence of discussion. Seeing the topics of the videos has some implications for what drives the most engagement. Videos about uncontroversial subjects such as cute animals have high views relative to their comments, suggesting that many people enjoy the content but don’t feel a need to discuss it. Contrast with videos about Skunk Ape (a cryptid animal) pandemics, and warnings from Stephen Hawking about Earth’s future habitability- controversial topics with comparatively high levels of active engagement.

The charts in this slideshow indicate the ratio of Likes to Dislikes for each video within the channel. This allows us to see the response to each video with only one response per viewer. Some interesting trends appear when examing the outliers. Positive videos, commemorations, and videos of cute animals and animal facts consistently have very high ratios, indicating that most of the reactions are “Likes.” In comparison, videos with more even or even negative ratios are often about controversial subjects like the Pandemic, animal combat, alarmist videos about the possibility World War 3, and strangely, sculpture instructional videos. Many of the highly disliked videos also make an appearance in the graphs of most commented- for example, the T.rex vs. Titanoboa (an extinct, giant snake) video, which is also the single most-watched video, also has a relatively high proportion of dislikes and number of comments. This suggests that, though it drove a lot of engagement, a lot of it was negative attention- arguments in the comment section and watchers disliking the video. It would be up to the channel owners to consider if this is a good trade-off.

Further Explorations/Limitations

The data scraper I used does not capture subscribers, so that was not included in the analysis. If I were to further iterate, I would make sure that I captured subscriptions in the dataset, as that is a vital metric that is highly valued by YouTube’s own algorithms. The exclusion of the National Air and Space Museum, the third most popular of the Smithsonian’s channels, by Youtube in the search is confusing and makes the drop-off from the most popular channels appear even more stark than it really is.

Improving my utilization of Tableau would also help. There as some visualizations I simply wasn’t able to figure out how to do such as area graphs and mouse-over explanations of data points with the data I had in the timeframe available to me, which was especially disappointing as the bubble graph of data breaches was my inspiration for this lab.