ERGM - size of the ergm term

Hi, I am trying to run an ERGM but the size of coefficient for the edges term is super big. I know there is something wrong but have no idea of how to find or fix the error. Will appreciate all your help
Below is the summary of my model:

ergm(formula = IL2013.n1 ~ edges + absdiff("age") + gwdsp(0.5,
fixed = T) + gwesp(0.5, fixed = T) + nodecov("rating") +
absdiff("rating") + nodecov("age") + nodefactor("independent"))

Iterations: 2 out of 20

Monte Carlo MLE Results:
Estimate Std. Error MCMC % z value Pr(>|z|)
edges 203.151858 1.788915 100 113.562 < 1e-04 ***
absdiff.age -0.061725 0.029950 0 -2.061 0.039312 *
gwdsp.fixed.0.5 -0.506044 0.203035 0 -2.492 0.012689 *
gwesp.fixed.0.5 0.744838 0.214677 0 3.470 0.000521 ***
nodecov.rating 0.150849 0.152178 0 0.991 0.321556
absdiff.rating 0.057478 0.166563 0 0.345 0.730035
nodecov.age -0.051442 0.000531 45 -96.873 < 1e-04 ***
nodefactor.independent.Y 0.714121 0.345360 0 2.068 0.038663 *

Signif. codes: 0 ‘’ 0.001 ‘’ 0.01 ‘’ 0.05 ‘.’ 0.1 ‘ ’ 1

 Null Deviance: 1081.3  on 780  degrees of freedom

Residual Deviance: 277.2 on 772 degrees of freedom

AIC: 293.2 BIC: 330.5 (Smaller is better.)

From the vignette:

To examine model diagnostics and check for degeneracy, use the mcmc.diagnostics() function.

Without a reprex that includes representative data it's not possible to say more.

1 Like


Have you checked the density of your network? Can you post the output of a call to summary: summary( IL2013.n1 ~ edges)

An edges term this large suggests either you have a very dense network, and/or that the edges term is compensating for a large negative coefficient on some other covariate. Depending on the values of the age nodal attribute, e.g., if the values are large, the negative coef on nodecov(age) could be driving the edges coef up.



when i am running this command
filter(mpg, year==2002)
i got this error
Error in filter(mpg, year == 2002) : object 'year' not found
what is the issue?
Is my RStudio is currpt???

Thank you very much, Martina! You were totally right. The network is not very dense (0.047) , so the problem was with the age variable. When I removed the nodecov (age) term from the model the size of the edges term goes down to -5, which I think is something normal, right? Is there a rule of thumb for a relatively normal coefficient for edges?


is there a year?

Glad this helped.

There's no rule of thumb -- it's both density and model specific. A coef value of 0 means that logit(p(tie)) = 0, and implies density is 50%. Most social nets have density well below 50%, so edge coefs are generally negative.

For age, you might consider centering the value of the variable on 0. If it's a substantively important covariate, it should be included in the model. And it's also good to include the "main effect" of age if you have the "interaction effect" of absdiff(age) in the model.



Thank you so much, this is really helpful. I will follow your advice and center the age variable to understand the main effect. I really appreciate your help :blush: :pray:t2:

Welcome :slight_smile:

To make sure questions like this get posted where statnet folks see them you might want to join the statnet_help listserv. Info available on our website:

The statnet team doesn't monitor other sites routinely ;).


Thank you so much, Martina!!!!
I will join I appreciate very much your advice :grinning:

1 Like

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.