Use of degcor ergm term with target statistics

[smjenness]

Hi @krivit : could you provide guidance on how to implement an ergm with target statistics for the degcor term. The help documentation for this term is not detailed.

For example, in the following edges-only model, I find that the mean degree correlation, output from degcor, is -0.07810538.

nw <- network.initialize(1000, directed = FALSE)
g <- ergm(nw ~ edges, target.stats = 1250)
g <- simulate(g, nsim = 1000, monitor = nw ~ degcor, output = "stats")
colMeans(g)

Then, trying to fit an ergm with this expected value, I get an error as follows. Is ergm expecting a matrix here, despite the model outputting a single correlation value?

g2 <- ergm(nw ~ edges + degcor, target.stats = c(1250, -0.078))
Error in sginv(cov(sm.prop), tol = .Machine$double.eps^(3/4)) : 
  Matrix a has negative elements on the diagonal, and snnd=TRUE.

Thanks,
Sam

[krivit]

The problem appears to be that degcor is undefined on an empty network, and its summary statistic ends up returning 0/0 or NaN. I guess we can treat an empty network as a special case that returns 0 by definition, or make it a parameter.

I am not sure it's a good idea to use it in modelling, either, since it lacks locality when its alternatives (degcrossprod?) don't.

[smjenness]

Thanks. Note that degcrossprod also generates the same error on an empty network. If we wanted to use it in the context of an egocentric/target stats approach, would we first simulate a random network with some mean degree, and then use that as the nw on the left hand side of the ergm forumula. Thanks for your help.

[krivit]

Thanks. Note that degcrossprod also generates the same error on an empty network. If we wanted to use it in the context of an egocentric/target stats approach, would we first simulate a random network with some mean degree, and then use that as the nw on the left hand side of the ergm forumula. Thanks for your help.

Huh... I had forgotten that degcrossprod divides by the total edge count as the last step. I wonder what would happen if we didn't?