[statnet_help] How does ERGMs deal with parameters that cannot be estimated from the data?

David Kretschmer dkretsch at mail.uni-mannheim.de
Mon Oct 12 08:23:17 PDT 2020

Hello dear list users,

I estimate an ERGM that includes a parameter that cannot be estimated from the data, i.e., it contains a dyadic covariate that has the value zero for each dyad.

To my surprise, running an ERGM on this data does still provide a coefficient estimate for the parameter that cannot be estimated from the data. Under normal conditions, the coefficient estimate is zero, and the standard error and all other elements in the ERGM output are NA. Still, I wondered why a coefficient value of zero is reported even though the parameter clearly cannot be estimated from the data.

Furthermore, when using the “constraints” argument of ERGM (constraints = ~bd(maxout = 5) in my case), the reported coefficient estimate is no longer zero but some other, seemingly random value that frequently is very high (e.g. 51 in one example run). Here, I also wonder how the ERGM arrives at this value.

In general, it is clear to me that these coefficient estimates are without substantive meaning. However, I wonder why they are still reported in the ERGM output and how the corresponding values come about.

Any help would be greatly appreciated.


David Kretschmer
Universität Mannheim
Mannheimer Zentrum für Europäische Sozialforschung (MZES)
A5, 6
68159 Mannheim
Tel.: +49-621-181-2024

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman13.u.washington.edu/pipermail/statnet_help/attachments/20201012/cb15fec1/attachment.html>

More information about the statnet_help mailing list