Repository navigation
nwdynam 1.6.0
Added
- Time-varying covariates, as in goldfish (attribute change events linked to the nodes): option
change(variable network [variable network ...])lets a covariate change over the log. The changes are an event network declared withnwset(one actor per event, sender = receiver, the new value as the event value); the dataset holds the values at the start. Optionincrementreads the event values as increments (goldfish:incrementinstead ofreplace).same(),diff(),sim(),ego(),alter(),egoalterint()andtertius()use the value that holds at the time of each event in the choice, rate (also withintercept) and choice_coordination sub-models. A change takes effect for the events after it; an event at exactly the time of a change still sees the old value (goldfish processes the events of a time point before the other event lists). Withinterceptthe observation period runs from the first to the last event of either list, and every change splits the time between two events, as goldfish does. - Estimates, standard errors and log likelihoods agree with goldfish 1.6.12 to 1e-6 on 42 random event logs with changes (the script and data are in
dev/, the test iscscripts/test_nwdynam_timevarying.do).
Changed
- The plugin protocol is 3: plugin binaries of 1.5.1 and earlier are recognised as too old (r(3495)). Until the build workflow has rebuilt the Linux and Windows plugins those platforms stop with r(3495). Fits without
change()give the same results as before to round-off and are not slower.
Tested on macOS with Stata 17 BE, Stata 18 BE and StataNow 19.5 MP (the full test suite; the goldfish cross-checks including 42 random logs with changes). Windows and Linux: not yet tested; the plugins are built by the build workflow, and until they are rebuilt those platforms stop with return code 3495.