Diss netstats: should they be -Persist or StartingNet-Persist?

statnet / tergm

Fit, Simulate and Diagnose Models for Network Evolution Based on Exponential-Family Random Graph Models

Other

27 stars 8 forks source link

Diss netstats: should they be -Persist or StartingNet-Persist? #55

Open martinamorris opened 3 years ago

martinamorris commented 3 years ago

it looks like the current implementation is Diss() = -Persist().

i get that this is correct in the sense that produces the right coefficient.

but the Diss() netstat is not what one might expect: the number of ties that dissolved.

we probably need to make this clear in the documentation, and point out that the tEdgeDissolution() function in tsna can be used on a nD object to calculate the dissolved ties.

krivit commented 3 years ago

In principle, it may be possible to redefine the Diss() operator to compute the decrease in the statistic in the post-dissolution network relative the previous time step's network. The parameter estimates, which depend only on change statistics, shouldn't be affected.

Should Form() then be redefined as the increase?

martinamorris commented 3 years ago

i'm not quite sure how to interpret your suggestion. sticking with the simple edges-only model:

the STERGM terms have always been the union (Form) and intersection (Persist), so we can't just change the definition, right?
the number of new edges from t0 to t1 is a different but related statistic (Form - edges(t0))
the number of dissolved edges is also different (edges(t0) - Persist)
the changestats are dyad level stats, not network level

i can explain all of these, and their relationships, but i didn't think you could just redefine the network level stats at will.

krivit commented 3 years ago

Sort of. Since the previous time step's network is conditioned on, shifting the Form statistic or what have you by the previous network's statistic will not change the probability model. That said, I certainly don't think we should be making that change at this stage.