There is no general theory of networks

February 16, 2018 9:40 AM

Scant Evidence of Power Laws Found in Real-World Networks. Remember when scale-free networks were all the rage? The idea was that real-world networks (including but very much not restricted to social networks) are surprisingly-often well-characterized by a power-law distribution: a node has k connections with a probability proportional to some (negative) power of k. This had implications for every science involving things connected to or interacting with other things! Well, about that...
posted by a snickering nuthatch (12 comments total)

Clauset uses his 2007-style analysis, which does not necessarily rule out fractal structure. Comparisons are to lognormal, Weibull, stretched exponential, which are all amenable to fractal structure.
posted by hleehowon at 9:47 AM on February 16, 2018 [1 favorite]

(I think that this thing may be better off as a rhetorical attack on Barabasi and his clones, which I guess the magazine thing notes)

Good to see Clauset getting press. Frequent collab CR Shalizi has the best statistical blog on the internet.
posted by hleehowon at 9:51 AM on February 16, 2018 [2 favorites]

“I think it’s pretty common for physicists who are trained in statistical mechanics … to use these kinds of analogies for why their model shouldn’t be held to a very high standard.”

daaaamn
posted by logicpunk at 10:19 AM on February 16, 2018 [4 favorites]

Clauset uses his 2007-style analysis, which does not necessarily rule out fractal structure.

"Fractal structure" != power law distribution. And the claims that were being made were specifically that this that and the other had a power law distribution, and even further that this implied specific mechanisms of organization.
posted by PMdixon at 10:38 AM on February 16, 2018 [1 favorite]

hleehowon : I would have guessed that Shalizi would have been an author on the paper just based on the FPP title. It fills out all of the Shalizi criteria: 1. About network theory and statistics 2. Smacking down statistically dubious claims.
posted by phack at 11:03 AM on February 16, 2018 [3 favorites]

Well, Clauset has a blog, but it's much less funny
posted by hleehowon at 11:08 AM on February 16, 2018 [1 favorite]

there is still plenty of good work going on regarding network modeling, including the aforementioned Shalizi’s work with Dana Asta on hyperbolic-geometry latent space embeddings.

just, none of it claims to have found a universal law.
posted by vogon_poet at 11:32 AM on February 16, 2018 [2 favorites]

An old post by Shalizi on treating power law models with appropriate rigor.
posted by a snickering nuthatch at 11:36 AM on February 16, 2018 [6 favorites]

Clauset uses his 2007-style analysis, which does not necessarily rule out fractal structure. Comparisons are to lognormal, Weibull, stretched exponential, which are all amenable to fractal structure.

100 percent thought this comment was generated via Markov chain. Now, time to RTFA!
posted by BlackLeotardFront at 12:37 PM on February 16, 2018 [3 favorites]

I'm pleased to see scale-free networks getting a bit of a smack. There was a period a 10-15 years ago when it seemed like almost every paper in the more computational side of biology that crossed my desk for review concluded with "... and it has the structure of a scale-free network!" (There wasn't actually an exclamation mark, but you could tell that they felt that it should be there.) After a while I ran out of polite ways to conclude my review with "So?"
posted by drnick at 3:24 PM on February 16, 2018 [5 favorites]

An old post by Shalizi on treating power law models with appropriate rigor.

Man, I wish I'd had this when I was in grad school researching the electron transport properties of certain materials; I always had a suspicion that the models our group was using were probably broken in ways that would remain totally invisible as long as we only kept using the calculations we had been, but I didn't have the right training to verify (or falsify!) that intuition in any sort of rigorous way.

Bookmarked for the next time I get into a fight with an economist, though.
posted by solotoro at 7:28 AM on February 17, 2018 [4 favorites]

Having recently done some gene network analysis myself, I remain fascinated about networks and dubious that they make sense everywhere they’re used.
posted by wintermind at 4:51 PM on February 18, 2018 [1 favorite]