How to specify an edgecov using a matrix.

[benrosche]

Hi!

I'm trying to run an ergm.multi model with an edgecov term but can't find the correct way to specify an edgecov using a matrix. I prefer the matrix approach because it's faster for large networks.

library(ergm.multi)
data(Goeyvaerts)

g <- Goeyvaerts[1:5]

# 5 ec_matrices that represent an edgecov 
ec_matrices <-
  purrr::map(g,\(x) { 
    n <- network.size(x)
    m <- matrix(rnorm(n*n,0,1),n,n)
    return(m)
  })

# 5 network objects that represent an edgecov
ec_nets <-
  purrr::map(ec_matrices,\(x) { 
    network(x, directed = FALSE, ignore.eval = FALSE, names.eval = "weight")
  })

# works
ergm(
  Networks(g) ~ edges + edgecov(Networks(ec_nets), "weight")
)

# does not work
ergm(
  Networks(g) ~ edges + edgecov(Networks(ec_matrices))
)

Thanks!

[krivit]

While in principle, you can treat the data structure returned by Networks() as a block-diagonal matrix, it is generally better to use the N(), which takes care of the bookkeeping and handles things consistently. Here's how I would do it:

library(ergm.multi)
data(Goeyvaerts)

g <- Goeyvaerts[1:5]

# 5 ec_matrices that represent an edgecov 
ec_matrices <-
  purrr::map(g,\(x) { 
    n <- network.size(x)
    m <- matrix(rnorm(n*n,0,1),n,n)
    return(m)
  })

# attach each covariate matrix to its respective network
for(i in seq_along(g)) g[[i]] %n% "weight" <- ec_matrices[[i]]

summary(Networks(g) ~ N(~edges + edgecov("weight")))
#>          N(1)~edges N(1)~edgecov.weight 
#>           28.000000           -2.354922

I hope this helps. If not, feel free to reopen.