Dyadic Imbalance in Networks

tags: [ src:paper , networks ]

src: link

• Actually does Local Testing on Graphs, as I’ve wanted to do. But in a very simple manner. Basically, you define dyadic imbalance as the fraction of cycles that are unbalanced. Cool.
• they also basically handle multiplex networks by essentially merging everything into one graph. Cool.
• data
  • what is sort of interesting is their application to IR is quite extensive (and very different to the social network setting)
  • they also have something related to voting decisions in the US Congress, which seems pretty interesting (like Wikipedia admin voting)
• method
  • they use the permutation test (which they explain in a very roundabout way)
  • they use something called a Heckman Selection Model, which I highly doubt solves the problem, but maybe it does?
    • this is some model that claims to fix sample selection biases
  • basically, they treat it like a covariate. but this is where we might be able to show something bad about their model. Basically, just like the whole spurious test when running linear regression on time series, doing node covariate regression, especially when you have graph statistics that are highly dependent is going to screw you over