nwdynam
Dynamic network actor models (DyNAM) for Stata: who acts next, and whom do they choose, in a stream of timestamped events.
Install Start with Tutorial 1 nwdynam on GitHub
What nwdynam is
nwdynam fits dynamic network actor models (Stadtfeld and Block 2017) to relational event data: a
list of who did what to whom, and when (phone calls, e-mails, messages, treaties). The model splits each
event into two decisions. The rate model says which actor acts next (some people call more often
than others, and more so when they are well connected). The choice model says, given the actor who
acts, whom the actor picks (people call those they called before, those who called them, …). Each
sub-model has its own effects and its own coefficients. nwdynam is a port of the R package
goldfish to Stata and is part of the nwcommands family of packages.
Who it is for
nwdynam is for people who work in Stata, or prefer it to R, and who have a stream of timed events
between actors. nwdynam follows the R package goldfish and is not a replacement for it: goldfish is the reference
implementation and the more complete program, and the page nwdynam and goldfish says what nwdynam has and what
it lacks. nwdynam is not affiliated with or endorsed by the authors of goldfish; errors in nwdynam are
the author’s. For how a DyNAM differs from the other longitudinal models of the nwcommands family, see
Intro to Longitudinal Network Models; the sister package for
the simpler relational event model is nwrem.
Key facts
| Sub-models | choice, rate (ordinal partial likelihood, or with an intercept the continuous-time hazard), and choice_coordination for undirected events |
| Effects | inertia, reciprocity, in- and out-degree, transitivity and other closure effects, covariate effects (same, diff, sim, ego, alter and more), windowed and weighted versions, a separate static network |
| Networks | directed one-mode, undirected one-mode (coordination), two-mode (a narrow set of effects) |
| Agreement with goldfish | coefficients and log likelihoods compared with goldfish 1.6.12 on its bundled Social Evolution data and on small hand-built examples |
| Engine | a compiled plugin for the likelihood and gradient (macOS, Linux, Windows); the optimization runs in Mata |
| Speed | Faster than goldfish in our benchmark: 7.7 to 76 times for the original structural effects (compiled plug-in), 1.2 to 2.4 times for an effect set that runs in Mata, with the gain shrinking as the number of actors grows; see Speed compared with R |
| Version | 1.2.0 (26 October 2026) |
| License | GNU General Public License, version 3 |
In this section
- Install: core first, then
nwdynam; releases, updating, the plugin. - Tutorial 1: A first DyNAM: event data, the rate and the choice sub-model, reading the table.
- Tutorial 2: Effects and interpretation: covariates, closure, windows, comparing models.
- Tutorial 3: Limits and diagnostics: what a small event log can and cannot tell you.
- nwdynam and goldfish: what is the same, what differs, a translation table.
- How to cite
Links
- Source code and releases: github.com/thomas-u-grund/nwdynam
- Stata’s own documentation:
help nwdynam - The R package: goldfish on CRAN