epsian / nattswap

Simulate network attribute and edge swaps for significance testing.

nattswap

Overview

nattswap is a small package designed to create network simulations in R which randomly swap attributes or edges in a network. It is based on the statenet family of network packages. If you are using igraph networks, I suggest you translate them using the intergraph package.

nattswap has two main groups of functions, one for attributes and the other for edges. You can see a full example application below.

Attributes

• run_att_swap_sims() will take in a network object, and run the simulations. It can save the simulations it creates if you want them.
• combine_net_att_sims() will reformat the results of run_att_swap_sims(), and combine the results of multiple if you did not run all simulations at once. Because swapped_attribute comes after ... in this function, it must be explicitly defined.
• get_att_swap_sig() will report the difference between the real and simulated networks.
• nattswap_compare() will plot the results and show how the actual means of network metrics compare to simulations.

Edges

• edge_swap() will take a network object, and randomly "re-wire" the edges, swapping edge pairs.

Installation

To install this package, please use:

# install.packages("devtools")
devtools::install_github("epsian/nattswap")

Example Network

Here is an example of the package in action. We'll use the flomarriage data from the ergm package (part of statnet).

# Load required libraries 
library(statnet)
library(nattswap)

# Load in example data from the *ergm* package
data(florentine, package = "ergm")

# Make a categorical variable to swap
# We will split the number of priorates (seats on the civic council) into high and low groups
flomarriage%v%"power_cat" = flomarriage%v%"priorates" > 40

# Run the simulations
# Here I ask for 500 simulations of with 500 swaps each, on the newly created "power_cat" variable.
flo_sims = run_att_swap_sims(net_to_swap = flomarriage, attribute = "power_cat", swaps = 500, sims = 500, verbose = FALSE)

# Combine simulations
# This is useful if you ran multiple simulations batches using `run_att_swap_sims`
flo_combined_sims = combine_net_att_sims(flo_sims, swapped_attribute = "power_cat")

# Get the significance of the differences between groups of "power_cat"
flo_sim_sig = get_att_swap_sig(flo_combined_sims, "power_cat")

# Make histograms to show the above differences
nattswap_compare(flo_sim_sig, metric_name = c("degree", "evc"), swapped_att = "power_cat")

Known limitations

• Currently only works on networks with one component
• Only works for undirected networks
• Will always only return values for degree, eigenvector centrality, betweenness, normalized betweenness, closeness, max cohesion, and nestedness

To Do

• add examples to help files
• fix code to remove warnings about network's "%V%"

Attribution

Based on the process shown in: Yang, Yang, Omar Lizardo, Dong Wang, Yuxiao Dong, Aaron D. Striegel, David Hachen, and Nitesh V. Chawla. 2016. “Gender Differences in Communication Behaviors, Spatial Proximity Patterns, and Mobility Habits.” ArXiv:1607.06740 [Physics].

Based on 'statnet' project software (http://statnetproject.org). For license and citation information see http://statnetproject.org/attribution