I'm unclear on the exact methodology used from the article and I can't be arsed to read the entire actual paper but I see some enormous glaring holes in this study. Enough to make me consider it utter bunk.
It already starts at their subject group: "1,063 adult Americans, recruited via Facebook ads" --> All facebook users, all americans, with seemingly no check on their political affiliation, nor a check on how long they've had their google ad-sense profile for instance, and the sort of people that would notice and respond to an ad on top of that = Holy selection bias Batman!
The research method is also utter crap. They start on a random video and basically have the user select a random recommended video. That's not how the algorithm and human users work though. You start at a video that for some reason drew your interest and you keep clicking on the video that draws the most interest. This will already give a very different response and input to the algorithm than this random clicking would and I strongly suspect it would re-enforce the ideological leaning/echo chamber effect tremendously.
Put on top of that the simple fact that the researchers themselves determine what is left- or right-leaning with an opaque "new methodology" based on smegging REDDIT!!! (from the paper: "Specifically, we use the observed behavior of sharing YouTube videos in the domain of ideological subreddits to calculate each video’s ideology that appears on Reddit. This set of more than 50,000 videos with an ideology score are then used as training data for a natural language classifier, which is then used to predict the ideology of any video on YouTube.) And people wonder why proper engineers laugh at the social "sciences".
This is imho utter garbage and it is my opinion the Reg should be ashamed of publishing such an uncritical article on it.
Biting the hand that feeds IT
