Is the Filter Bubble Real?

Just a few short months ago, Eli Parser gave a TED talk about something he called the “filter bubble.” His thesis was that automatic personalization of search and related technologies meant that users would tend to see information that supported their general viewpoints and rarely if ever see anything that contradicted them. The truth may be rather different from this.

There’s a related phenomenon with an older name, as Alexander Zwissler reminds us: confirmation bias. The concept describes the all-too-human tendency to seek out information that confirms cherished opinions we already believe to be true, while dismissing any information that contradicts these closely-held beliefs. It’s one of the reasons there are so many different news-related TV channels and websites in the world.

Preventing confirmation bias is also the reason that one of the classes every would-be scientist is supposed to take in college covers proper experimental design, and another one covers statistics. When confirmation bias creeps into your science, bad things happen. Fortunately, you don’t have to be a scientist to test the phenomenon of filter bubbles, as Zwissler found.

In an article for the Contra Costa Times, Zwissler decided to put the filter bubble theory to the test. He and 10 of his friends performed an experiment. They were roughly alike in age, athletic ability and hobbies. They grew up together in Oakland, but they’d since scattered geographically – and politically. “We cover the spectrum from progressive left to conservative right and all points in between.” It might be far from a perfect scientific sample, but if the filter bubble exists, one would expect to see a sign of it.

Zwissler’s test was simple: all 11 of them would search Google for three specific topics at the same time. Zwissler’s 10 friends would then send him their links. The three topics chosen were “global warming,” “Oakland,” and “mountain biking.” Presumably, everyone had to perform the searches while they were logged into their individual Google accounts. (I say “presumably” because I couldn’t find anything from Zwissler in the article saying that explicitly, though Rob Young does state this in his coverage of the item on Search Engine Journal).

Zwissler expected to see a wide variety of results – climate denial sites from his conservative friends, while his liberal ones would see links to the Sierra Club, for example. Or maybe one group would get reports of crime in Oakland, while another would get the latest Uptown restaurant reviews. The actual results did not meet his expectations.

On the contrary, the terms he expected to yield the greatest differences in links between his friends showed only modestly different results. “Mountain biking,” on the other hand, showed much larger differences, both in the links shown in the actual results and in the ads displayed. What would cause this?

Well, there are several possibilities. The differences in mountain biking information stem from the differences in location. The similarities in the other results, however, could come from at least two possible causes which are not mutually exclusive. First, the differences between the political opinions of Zwissler’s friends aren’t as great as he believes. Second, the filter bubble effect is not as great as Parser believes.

Personally, I’d like to see more experiments before trying to read too much into this. The only result that seems clear to me is that there are geographically-based differences in what Google will show someone who is logged into their Google account. So I agree with Young’s advice in his piece on this topic: “…location is the biggest way the search page is shaken up. Take this as a signal tha tyour page your get involved in location-signaling whenever it’s appropriate to your message or brand.” Good luck!

[gp-comments width="770" linklove="off" ]