YouTube's White House Channel Gymnastics

Following the inauguration of Joe Biden and his administration in late January, his team has also overtaken the White House YouTube channel. The first video posted by the new administration was uploaded on January 26th. Since then, the volume has grown to almost 400 videos as of mid-April. These videos typically feature White House press conferences, COVID-19 updates, and various other official meetings and events.

Over time, many on the internet started noticing dubious developments on the YouTube channel. The number of dislikes on many videos appeared to sometimes decrease. Without outside interference, the only way the dislike count could go down would be when a viewer, who had previously disliked the video, subsequently decides to ‘un-dislike’ it. Though that can sometimes happen, not many people choose to return to a previously watched video, and even fewer tend to then change their mind about upvoting or downvoting it.

Thus, the unnatural-looking decreases started to attract more and more attention, to the point of now being closely tracked by a standalone project set up just for that purpose. The 81m.org website aims to question the widely-promulgated narrative that President Biden is popular with the broad public by pointing to ‘manipulation’ with the popularity indicators on YouTube. Such ‘manipulation’ is presumed by the site to be politically motivated, with the aim to present Biden as a President esteemed among the people.

The inspiration for the project was a January 27th article by Zoe Phin, who had plotted and analysed the like/dislike counts on several White House videos. She showed that at specific times of the day, the dislike count on those videos would decrease substantially. Save for an off-hand comment about Biden’s popularity, in her article, she did not speculate about the reasons for this, though much of her audience viewed YouTube’s involvement in the statistics as clearly political.

81m.org then began to compile the same statistics about all the videos on the White House YouTube channel. As part of its methodology, the website periodically looks at YouTube API data and logs the number of likes and dislikes on a given video, taking two sets of data - the ‘official’ numbers and the ‘real’ numbers. The latter is calculated simply by ignoring decreases in a video’s dislikes. Such a methodology might be marginally inaccurate (such as where some viewers themselves remove their dislike), but where large movements in the numbers are observed, those should point towards genuine interference.

The site then calculates the approval rates (the proportion of likes to the sum of votes), determining both the ‘official’ rate and the ‘real’ one.

In the following numbers, I took into account the videos before April 6th - the first 10 weeks of uploads since the first one on January 26th. This totals 331 videos. From these uploads, there were 254 (roughly 77%) in which the ‘official approval rate’ was at least twice as high as the ‘real approval rate’. For 99 videos (30%), the ‘official approval rate’ was more than five times the ‘real’ one.

Finally, 38 videos featured ‘official approval rates’ more than ten times larger than their ‘real approval rates’. Here are the numbers for each of those at the time of writing: 12.99, 19.09, 37.76, 11.13, 10.68, 12.26, 14.42, 12.72, 11.78, 18.02, 11.69, 10.27, 12.87, 23.14, 24.93, 22.42, 17.19, 28.10, 13.31, 13.59, 11.14, 13.82, 13.10, 16.01, 13.82, 10.91, 11.58, 11.57, 12.81, 11.79, 13.13, 10.88, 13.37, 15.05, 10.75, 16.88, 12.73, 10.41. 

Eighteen of those 38 top-interfered-with videos were related to COVID-19, and twelve featured Vice-President Harris.

Notably, the interference has increased over time. For the first quarter of the 331 uploaded videos, the average ‘official approval rate’ was 2.71 larger than the average ‘real approval rate’. By the fourth quarter, this number had grown to 7.68.

On a chart showing the approval rate, it is clearly visible how the ‘artificial’ part (orange) makes up the majority of most videos’ official approval rates.

The ‘manipulation amount’ metric used on the 81m.org website is different from the interference ratio calculated above. 81m.org uses the absolute numbers of dislikes removed from the videos and compares them with the dislikes that remain as the ‘official’ number. The ‘manipulation amount’ for each video, then, is the ratio between the absolute number of ‘real dislikes’ and ‘official dislikes’. According to this metric, the most-’manipulated’ video is from April 2nd, where the total number of dislikes has been decreased from 16,677 to 251, which makes the ‘manipulation amount’ a staggering 6544.22% (at the time of writing). The video in question features that day’s COVID-19 press briefing.

To illustrate that the issue of vanishing dislikes is specific to the White House channel and not present across the board on YouTube, 81m.org also features ‘comparison data’ where the same numbers are gathered for several other, non-political channels. For these videos, the “manipulation amount” generally stays near zero, only with a handful of exceptions. Often, that number is slightly below zero, reflecting the ‘natural’ growth of dislikes on a video over time and the likely absence of any interference from YouTube’s side.

Although it is often assumed that YouTube is manipulating the popularity statistics for political purposes, so far this has not been confirmed. In February 2019, The Verge reported on YouTube’s issue with people supposedly “weaponizing the dislike button.” At the time, several options were “lightly being discussed” to stop the “dislike mobs” from purposefully downranking videos in other users' search results. One possibility was to implement a “checkbox”, where those intending to press the dislike button would also have to select a reason why they don’t like the video. This has not been implemented. 

“The last option, which [Tom Leung, director of project management at YouTube] describes as the most extreme option, is just to remove dislikes entirely. It’s not ‘as democratic,’ according to Leung, as ‘not all dislikes are from dislike mobs,’” continues The Verge.

YouTube has not done away with the dislike button altogether. However, it seems to have resorted to the removal of some dislikes according to an unclear process or algorithm. This has been confirmed by YouTube multiple times. For example, “YouTube regularly removes any spam likes or dislikes from your videos,” says TeamYouTube on Twitter. Elsewhere, TeamYouTube reiterated: 

“We always validate the activities & legitimacy of accounts added on your likes/dislikes report. This is to make sure that our site metrics are free of spam.”

The latter Tweet features a link to “learn more,” which leads to YouTube’s ‘Help’ site. There, in the ‘FAQ’ section, likes/dislikes movements are addressed by a brief comment: “You may see like and dislike counts change as some may be marked invalid and periodically removed from the counts. Learn more about our Likes Policy.” The Likes Policy link leads to YouTube’s article on “Spam, deceptive practices, & scams policies,” where the only mention of likes or dislikes is a section of the rules which prohibits users from selling engagement metrics. That is where this particular trail ends, with inconclusive answers.

YouTube also spoke with PolitiFact about their handling of dislikes. The company confirmed that it had “systems in place to ensure that engagements with videos - such as likes and dislikes - are authentic, so that the analytics information it provides is reliable. Those systems worked as designed to remove spam engagements” on one of White House’s videos, which PolitiFact focused on. 

While YouTube admits to removing dislikes when it considers them to be a result of spamming rather than ‘genuine’ engagement, any political motives are difficult to prove. The sheer numbers of dislikes deleted from the White House channel - now numbering several million - seem to be indeed exceptional. It must be said that it is entirely possible that such an account would be much more exposed to attempts to manufacture unfavourable engagement statistics by nefarious actors with semi-expert coding skills. However, it is just as well possible that YouTube is adjusting the channel’s numbers to earn itself some political favours. This particular whistle is still waiting for someone to blow it.

Readers can find the source numbers for my own analysis here (including the charts). All the input data is taken from 81m.org on April 15th.