Remove base= arguments from ergm terms?

[krivit]

They've been deprecated (with a warning) since 3.9.4 (early 2019), and has been throwing warning messages since then. Similarly, sociality() (and only sociality()!) also has attribute arguments for homophilous sociality, which can now be better handled using F(~sociality(...), ~nodematch(...)).

@mbojan , @drh20drh20 , @CarterButts , @martinamorris , @sgoodreau , @handcock ?

[mbojan]

I guess 7 years is enough for evolution to exert its pressures on users... At the same time, the alternative might not be obvious to a lot of people (I bumped into miss-pecifying levels or levels2 myself more than once). Perhaps we should have a vignette demonstrating how to go ahead in life without base= and with all the other attribute-related arguments? The documentation is there, but I think it is very terse / compressed and often without examples.

[sgoodreau]

I agreew with @mbojan: these other arguments are a bit less intuitve for me, despite all of my familiarity with the packages. More documentation and a vignette would be very useful here. WIth those in place I can see removing them finally.

[CarterButts]

I can never figure out how the new terminology works, and probably still have a lot of code using `base`. So I would certainly second the call for a more approachable documentation solution. (I'm also not very fond of the trend towards having to use operator expressions, because they make `ergm` and friends *much* less hackable. It's easy to parse simple `+`-separated terms, but parsing some complex formula language is not feasible....)

…

On Tue, Feb 10, 2026 at 11:05 AM Steven Goodreau ***@***.***> wrote: *sgoodreau* left a comment (statnet/ergm#648) <#648 (comment)> I agreew with @mbojan <https://github.com/mbojan>: these other arguments are a bit less intuitve for me, despite all of my familiarity with the packages. More documentation and a vignette would be very useful here. WIth those in place I can see removing them finally. — Reply to this email directly, view it on GitHub <#648 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJM3GFIRMU5ZGFO4VSECFT4LISHVAVCNFSM6AAAAACUSH4SZGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTQNZZHE3TCMZWG4> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[krivit]

@mbojan , @sgoodreau , There is already a vignette about specifying nodal attributes, and we could probably add a section on converting from base=, and we could probably update each term's help individually.

levels matches R's semantics more closely than base, since it's simply indexing the vector of potential levels (which is why levels = TRUE includes all levels).

It's easy to parse simple +-separated terms, but parsing some complex formula language is not feasible....

If the problem is constructing formulas programmatically, you can use I(formula) to parse and insert whatever you want into the formula.

@CarterButts If not, can you give me an example of the kind of hackability you are looking for?