0

Can TERGM be used to fit bunch of networks files from 1.net to n.net? With each file, the number of nodes and vertices increases. I have

For example, when we fit a particular network (suppose 6.net), we run the following after preprocessing the network (i.e finding the metrics value and adding the metrics to the nodes)

network_fit<-ergm(my_network~edges+nodecov("degree")+nodecov("betweenness")+nodecov("closeness")+nodecov("pagerank")+nodecov("eigen_c")+nodecov("clustcoeff"))

Can the same be done using TERGM?

I have fit the network using TERGM over 49 networks (1.net to 48.net) fit <- tergm(net_y.dyn~edges + nodematch("degree"), estimate = 'CMLE', times = c(0:48))

I am getting the following warning

Starting maximum pseudolikelihood estimation (MPLE):
Evaluating the predictor and response matrix.
Maximizing the pseudolikelihood.
Finished MPLE.
Stopping at the initial estimate.
Evaluating log-likelihood at the estimate. 
Warning message:
In NetSeries(nw, times, NA.impute = control$CMLE.NA.impute) :
  Active vertex set varies from time point to time point. Estimation may not work.

Can anyone explain why this warning is showing?

Phil
  • 7,287
  • 3
  • 36
  • 66
Su1
  • 1
  • 1

1 Answers1

0

TERGM assumes the nodeset is fixed. The warning is because you are submitting a series of networks of different sizes.

Having said that, it is possible to incorporate exogenous node entry and exit processes into a TERGM -- such specifications are used in EpiModel framework, an example can be found here https://github.com/EpiModel/EpiModel-Gallery/tree/main/2018-08-SIwithVitalDynamics. A more general model with endogeneous node entry and exit is a subject of ongoing research.

Michał
  • 2,755
  • 1
  • 17
  • 20