Excellent paper.

TL;DR Majority illusion paradox occurs when many people are observing a small and loud group "saying" something that is actually globally rare. Imagine 100 people who all observe three people who say the same things.

Own a media company and get people to believe in junk. Find the most connected individuals among a group and make them "active". If an "inactive" person perceives .5 of their connections are active they may switch to the active state themselves.

The "type" of network can affect the perceptions.

Isn't that basically the phenomenon that all PR evangelists use to get an idea out?

Well, you can "say" something over a channel but that does not "create" the perception that it is popular. So, this paper describes how things are perceived as popular when they are not really popular in a "numerical" sense. When Johnny says "everyone" is doing it, what he means is that all of the people he listens too are saying that "everyone" is doing it. In fact, only 2 people are doing it.

And all successful activists.

Excellent paper? This seems like an observation one could make in the shower and publish as a tweet.

Agreed. It's basically the well known "vocal minority" idea just applied to graph theory. I did not learn anything new from reading it.

Not necessarily vocal, just well-connected.

It is uploaded to arxiv. It won't be published in a journal, since it is a trivial and non-novel result that _broadcasters_ have a larger audience than non-famous people.

As someone who has published a graph theory paper (in a journal), I doubt this is far away from being state of the art work.

A lot of people are trying to apply graph theory in wild situations where it doesn't belong, yet this basic work on social networks and social graphs hasn't been studied in an academic way.

I have a feeling Karl Rove and David Axelrod know this quite well.

This is really interesting. My interpretation of this, in practical terms, is that if one were to pay two celebrities to Tweet or make Instagram posts, the optimal strategy would be to find the pair that has the highest overlapping group of followers. It would be interesting to see if this would increase virality enough to justify the additional costs.

It's annoying how the article does not properly say what institution the research comes from in the very first line.

They attribute it to the University of California which consists of ten different campuses and typically Berkeley gets away with being called just University of California, or CAL, because it was the first.

However upon looking up the researchers' name it turns out that they belong to none of these and are actually from the University of Southern California.

I don't understand your point here. Is it that you don't like the sloppy journalism at phys.org? Or that you don't want to bother with reading pre-prints from researchers at schools that aren't Berkeley?

What? Of course it's sloppy journalism. Stating who did the research is important. It's less important in this situation since they link to the actual paper, but it's one of the first things you want to know when you do a report (Who, what, where...). Being incomplete is bad enough, getting it outright wrong is even worse.

Permanently sticky this to Reddit so they stop thinking they're a representative sample of the population at large.

Which of the private subreddits do you suggest they post it to :)?

Now imagine the power search engines have when they decide what content/news stories to show you. They could (have??) sway elections.

I'm more curious about, it's known that Facebook uses algorithms to display content based, in part on the popularity of that type/user's history of content popularity.

So to what extent does that create a self-fulfilling prophecy where people who tend to post content generally likable to their friends then have that content displayed around and liked more, having a self-reinforcing effect on how Facebook "feels" and how people understand what they are expected to post to Facebook.

Sort of a fast-foodization of thought?

“I live in a rather special world. I only know one person who voted for Nixon. Where they are I don’t know. They’re outside my ken. But sometimes when I’m in a theater I can feel them.” — Pauline Kael (this is what she actually said) http://www.vanityfair.com/culture/2012/10/The-Fraudulent-Fac...

This is why Nate Silver, who after all mostly does his work by taking polls as a whole seriously, continues to surprise some people: They live in a bubble. They don't see who the majority is actually supporting. They are actually surprised by election results, even if the polls have largely been pointing one direction consistently for months on end.

(Silver ignores some polls and applies correction factors to others. This is due to those polls' demonstrable idiocy and/or bias, as empirically demonstrated by how far they are from reality when the real events happen.)

Wang does pretty well, too, and his models are simpler: http://election.princeton.edu/

My point is, polls work, statistics is real, and numbers remain the primary way to determine which of two things is larger. Enter echo chambers if you want to, but don't live in one.

I don't understand why this is called an illusion. How about just counting the edges as a measure of popularity? Surely nothing "appears to be more popular" than it is -- if there are more outgoing edges to other nodes, it is more popular. Popularity has always been about popular people.

Because each node's view of neighboring nodes is a biased sample of the global data set.

A related example: the average person is less popular than their average friend, and therefore feels that they are below average popularity. This happens since popular people are weighted more heavily in the sampling, but people don't realize they are confusing the local average with global average.

Perhaps social networks should compensate for this effect, and show us the real world instead of a hyped world.

I'm not sure if facebook would be willing to do this, but perhaps GNU social?

What are you suggesting specifically?

If some item is shared (or liked) there is a probability p that you will see it on your newsfeed.

This probability p should be a function of the popularity r (in your circles) directly, and it should not depend on who is sharing the item.

this has seemed obvious to me since I first got on computer networks in the 80s. it's nice to have my intuition be scientifically validated. I don't think there will be much impact from this on a human level, though. maybe some marketers who didn't really see the formula before will benefit.