NEWS | R Documentation |
NEWS file for the ergm.ego
package
Changes in version 1.1.1
NEW FEATURES
-
An
EgoStat
forabsdiffcat()
has been added.
BUG FIXES
-
degreedist()
has been fixed.
OTHER USER-VISIBLE CHANGES
-
logLik()
method forergm.ego
objects has been added; it produces an informative error message. -
Documentation fixes.
Changes in version 1.1.0
NEW FEATURES
-
ergm.ego()
now has abasis=
argument. So doessimulate.ergm.ego()
, for consistency (as an alias for thepopsize=
argument). -
simulate.ergm.ego()
'spopsize=
argument can now be a network object, enabling simulation from any starting network.
BUG FIXES
-
gof.ergm.ego(GOF="degree")
now handles the case in which the observed or simulated degree distribution is dense and the LHS network is small more gracefully.gof.ergm.ego()
was scrambling the the order of ESP terms.simulate.ergm.ego()
is now more robust to models with offsets and extreme “dropped” statistics.ergm.ego()
(viacontrol.ergm.ego(ppopsize=)
) andsimulate.ergm.ego(popsize=)
can once again takedata.frame
s andtibble
s to specify the pseudopopulation network composition directly.
OTHER USER-VISIBLE CHANGES
-
simulate.ergm.ego()
now preserves some of the attributes attached bysimulate.ergm()
to the statistics matrix, including"monitored"
. -
simulate.ergm.ego()
no longer supportsergm.ego
objects fit under under ergm < 4.
Changes in version 1.0.1
BUG FIXES
-
Documentation fixes, particularly for compatibility with ergm 4.2.
-
Summary for
ergm.ego
fits now displays the original call rather than the instrumentalergm()
call.
OTHER USER-VISIBLE CHANGES
-
control.ergm.ego()
praameterignore.max.alters=
now defaults toTRUE
, since simulation studies (Krivitsky, et al. 2020) showed that they did more harm than good.
Changes in version 1.0.0
NEW FEATURES
-
This package now uses the egor package's
egor
class for data storage and manipulation. A converteras.egor.egodata()
is provided. -
ergm.ego()
now supports complex survey designs set onegor
objects. -
ergm.ego()
and the summary methods can now fit triadic effects (gwesp
,esp
,transitiveties
) when alter-alter ties are available. -
ergm.ego()
can now handle missing alter attributes in some circumstances, and provided they are missing completely at random. -
A number of new egostats have been implemented, including
gwdegree
-
A number of improvements to the goodness-of-fit routines.
-
snctrl()
UI for specifying control parameters is supported. -
Curved ERGMs are now supported; this capability should be considered experimental, as uncertainty estimates have not been rigorously derived.
-
For nonscaling statistics such as
meandeg
, standard errors can now be computed. -
Network size adjustment can now be disabled during fitting.
BUG FIXES
-
Various fixes to
degreedist()
,mixingmatrix()
, and other methods.
OTHER USER-VISIBLE CHANGES
-
The function that was previously
as.network.egodata()
for constructing an empty network having the same composition as the egocentric dataset has been superseded bytemplate_network()
. -
Manually specified pseudo-population is handled better.
-
degreedist()
method for egocentric data now defauts to not making plots. -
mixingmatrix()
method for egocentric data now returns atable
.
Changes in version 0.6.0
NEW FEATURES
-
predict.ergm.ego
, apredict
method forergm.ego
has been implemented. (Thanks, MichaĆ Bojanowski.) -
Nonscaling statistic
meandeg
has been added.
OTHER USER-VISIBLE CHANGES
-
EgoStat.*
functions no longer need to be exported, reducing namespace pollution.
BUG FIXES
-
ergm.ego
now detects when a coefficient has been dropped byergm
due to the statistic having an extreme value and subsets the variance matrices accordingly. -
control.ergm.ego
now callsmatch.arg
onppopsize
only ifppopsize
is of classcharacter
. This allowsppopsize
to be of classnetwork
when callingcontrol.ergm.ego
. -
A more thorough search mechanism for
EgoStat.*
functions no longer requires them to be exported.
Changes in version 0.5.0
NEW FEATURES
-
ergm
's new nodal attributes user interface has been extended toergm.ego
. -
mixingmatrix.egodata
anddegreedist.egodata
now have an option to ignore sampling weights. -
Simulation frmo an
ergm.ego
fit now inherints the constraints. -
It is now possible to specify the (pseudo)population network temlate directly by passing it to
control$ppopsize
. -
It is now possible to infer main effects (
nodefactor
andnodecov
) when the attribute has only been obseved on the egos.
BUG FIXES
-
A wide variety of minor bugs has been fixed. See commit log and issue tracker for details.
OTHER USER-VISIBLE CHANGES
-
A number of robustifications have been made.
-
ergm.ego
now produces sensible error messages when terms have alter categories that egos do not. -
Chad Klumb has been added as a contributor.
-
gof.ergm.ego
's default MCMC.interval is now the MCMC.interval of the ergm fit scaled by the ratio between the fit'sMCMC.samplesize
and GoF control'snsim
. -
gof.ergm.ego
now only calculates GOF for degree values up to twice the highest observed in the data or 6, whichever is higher with an additional category to catch the higher values.
Changes in version 0.4.0
NEW FEATURES
-
mm
term has been implemented.degreedist
now has an option to not plot, and returns the calculated degree distribution (invisibly, if plotting). -
offset
terms are now handled. More
EgoStat
now handle more options that theirergm
counterparts do.-
ergm.ego
'sppopsize
control parameter andsimulate
method forergm.ego
'spopsize
argument now take a data frame of egos to use as the pseudopopulation.
BUG FIXES
Package now works with
ergm
3.9.-
degreedist
now handles sampling weights correctly, and has been fixed in other ways. Bootstrap and jackknife now handle one-dimentional stats correctly.
-
mixingmatrix.egodata
now handles ego ID column names other thanvertex.names
. Thanks to Deven Hamilton for reporting this bug. Non-numeric ego IDs are also handled correctly. -
mixingmatrix.egodata
no longer rounds the row probabilities before returning when calledrowprob=TRUE
.
OTHER USER-VISIBLE CHANGES
-
degreedist.egodata
is now anegodata
method ofdegreedist
.
Changes in version 0.3.0
NEW FEATURES
This is the initial public release.