Significance Tests R

Very new to Social Network Analysis and R (read a few introductions online, created a few graphs on my own) but I am very eager to learn different ways to apply it. That said, what are common significance tests applied to networks, and more specifically, if two 开发者_开发技巧network graphs are "statistically different?"

Any help/tutorials you can point me to - especially in R - will be very much appreciated!

Thanks,

First, the Graphical Models in R Task View, (on CRAN) lists probably every R Package that is primarily related to graph theoretics or any application like social network analysis.

(Task Views are just collections of Packages related either by domain/application (e.g., Finance) or by technique (e.g., Regression)).

If you are fairly new to network modeling, and you don't have a background in graph theory, let me suggest the R Package netmodels (the value is not limited to novices though). netmodels is in essence a set of wrappers around lower-level functions in a C library called igraph, which has R bindings (also called igraph).

So for instance, using netmodels you can just call the function

do.base.description

on your network and it will return a list several key network parameters and that network's value for each (for instance, number of nodes, number of edges, average degree, average path length, etc.)

So by using netmodels you can quickly get a high-level description of your network without having to calculate those high-level descriptive parameters yourself from 'articulation points the basic graph theoretic components like 'articulation points', 'eigenvectors centrality', and 'Cocitation coupling'.

Three other similar R Packages which wrap lower-level graph theoretic functions are inetwork, RSiena, and sna.

Regarding "significance tests" for networks. I believe that analysis is directed to the statistical significance of key parameters/attributes of the networks under study.

So what are those key parameters? Perhaps there's a consensus on this, but more likely it depends on the particular problem you are trying to solve. Here's my list of key network parameters i nearly always calculate (i'll try to list a function and its R Package next each attribute):

• cumulative degree distribution (count its edges, or links, for every node in the network, let's you estimate nodes-to-edges ratio) R Function: plot.c.degree in package netmodels

• Krackhardt connectedness (returns a score from 0 to 1) R Function: connectedness in package sna

• average path length (shortest directed or undirected) path between nodes R Function: average.path.length in package igraph