edges between groups

[rwss2021]

I realized this package could be really useful to my current research, but I have a general question on this.
I noticed that the term nodematch() accounts for # of edges within the same groups; however, if I wanted to see whether edges between different groups are significant in the network or not, which function am I supposed to use? or do I need to write a customized one on this?
Thank you for your time!!!

[sgoodreau]

Hello - the term that allows for more generalized modeling of differences in mixing patterns among categories is nodemix.

[rwss2021]

Thanks a lot!!

[mbojan]

As usual there is more than one way to go about it:

• nodemix as Steve suggested is the most general of the ready-made terms which seems appropriate
• You can create one or more dyadic covariates, each a matrix with 0s for within-group dyads and 1s for dyads between the groups of interest and use it/them with dyadcov or edgecov terms. The interpretation of the coefficients depends on what set of such created covariates you include in the model, similarly to treatment contrasts in linear models.
• If you are just interested in the between- vs within- then you can simply look at (opposite of) nodematch because if a tie is not within-group it must be between-group
• If you have only two groups and code the vertex attribute identifying these groups to have values 0 or 1 you can use absdiff (will result with 0s for within-group dyads and 1s for betwen group dyads)

[martinamorris]

I'll also note, in case this was your question, that if you want to test whether there is a statistically significant difference in the presence of within-group ties -- vs. the null hypothesis of random mixing -- that's exactly what the ergm function is built to do. You estimate the coefficient on the term of interest (nodematch, or the terms noted above), and examine the p-value of the test statistic, which is shown in the usual summary table of the fit object.