Gorka Zamora-López

Analysing and interpreting data can be a complicated procedure, a maze made of interlinked steps and traps. There are no official procedures for how one should analyse a network. As it happens in many scientific fields the “standard” approach consists of a set of habits that have been popularised in the literature – repeated over-and-over again – without always being clear why we analyse networks the way we do.

Imagine we wanted to study an anatomical brain connectivity made of $N = 214$ cortical regions (nodes) interconnected by $L = 4,593$ white matter fibers (a density of $\rho = 0.201$). Following the typical workflow in the literature we would start the analysis by measuring a few basic graph metrics such as the degree of each node $k_i$ and their distribution $P(k_i)$, the custering coefficient $C$ and the average pathlength $l$ of the network. Imagine we obtain the empirical values $C_{emp} = 0.497$ for the clustering and $l_{emp} = 1.918$ for the average pathlength.
Continue reading Network analysis abuses of null-models →