Differences in centrality measures between igraph and tnet

igraph, r, social-networking

Solution

After posting this question, I stumbled upon a blog maintained by Tore Opsahl, maintainer of of the `tnet` package. I asked this same question of Tore using the comments on this post of the blog. Here is Tore's response:

Thank you for using tnet! `igraph` is able to handle weights; however, the distance function in `igraph` expects weights that represent 'costs' instead of 'strength'. In other words, the tie weight is considered the amount of energy needed to cross a tie. See Shortest Paths in Weighted Networks.

Thus, if you run the following code provided by Tore (which takes the inverse of the weights before passing them to `igraph`), you obtain equivalent closeness scores for both `tnet` and `igraph`.

> # Load packages
> library(tnet)
>   
> # Create random network (you could also use the rg_w-function)
> m <- expand.grid(from = 1:4, to = 1:4)
> m <- m[m$from != m$to, ]
> m$weight <- sample(1:7, 12, replace = T)
>   
> # Make tnet object and calculate closeness
> closeness_w(m)

     node closeness n.closeness
[1,]    1 0.2193116  0.07310387
[2,]    2 0.3809524  0.12698413
[3,]    3 0.2825746  0.09419152
[4,]    4 0.3339518  0.11131725

>   
> # igraph
> # Invert weights (transform into costs from strengths)
> # Multiply weights by mean (just scaling, not really)
> m$weight <- mean(m$weight)/m$weight
> # Transform into igraph object
> igraph_g <- graph.data.frame(m)
> # Compute closeness
> closeness(igraph_g, mode = "out")

        2         3         4         1 
0.3809524 0.2825746 0.3339518 0.2193116

Problem

Original source