[statnet_help] node entry and exit in TERGMs

Leifeld, Philip philip.leifeld at essex.ac.uk
Tue Aug 18 11:56:49 PDT 2020


Hi Holly and other colleagues,

In my view, how composition change of the node set is dealt with should
depend on the nature of the composition change. It should depend on two
conditions: (1) Is the presence or absence of a vertex a function of
prior edges or vertex attributes? And (2) does the presence or absence
of a vertex cause or suppress the presence or absence of other vertices
(or edges in dyads not involving the focal vertex) at later stages? If
the answer to both questions is no, the composition change and the
network dynamics are mutually independent processes, and a simple
procedure for taking out the nodes at the respective time point (or
between time points) should do the trick. In the TERGM case, counting
subgraph products that involve nodes at both t and t-1 will then simply
not add anything to the respective statistic. The btergm package, which
builds heavily on the ergm package, contains procedures to either remove
nodes that are not present or, alternatively, include them and model
ties to or from them as "not allowed" using structural zeros. The
package reports details of vertex composition and how this was dealt
with before estimation. See details here:
http://dx.doi.org/10.18637/jss.v083.i06. Similarly, the SAOM literature
and implementation contain ways to make the simulations unaffected by
small amounts of composition change, as per the article recommended by
Christian, such that only active actors are allowed to change their
outgoing relations with other active actors at any time, although the
SAOM simulations often break down when you have extensive composition
change on the order of 40 percent of the nodes per time point or so. If
the answer to the two questions above is yes, then some more elaborate
procedure is needed, and I think Carter's and Zack's excellent articles
will be very helpful starting points in the TERGM case.

The Czarna et al. paper used an ERGM for 15 independent cross-sectional
networks and two time points with around 40 percent composition change
overall and, in some cases, going up to 100 percent composition change
in a couple of networks. This makes a SAOM application infeasible
because the simulations would quickly break down. But irrespective of
this, modeling this system as a temporal process would have meant using
t=1 as a lagged dyadic covariate for t=2, which would have been
undesirable. This would have made answering the research question of how
psychological traits matter for one's position in a network in a
first-impression context versus in networks later in the year impossible
to answer, by design. Incidentally, it would have also removed 70
percent of the dependent variable by limiting the observations on the
left-hand side of the equation to the available nodes and only at t=2.
So I think in this case modeling the temporal variation in covariate
effects on network position using temporal dummies and interactions of
the covariates with time is a relatively minor assumption compared to
the alternatives. But YMMV, and applied research is always a collection
of assumptions people are making in the research process. It's important
that they are transparent so we can build on each other's work in
sensible ways.

I think network dynamics are an exciting problem (or opportunity) to
work on, and much is yet to be learned. Advice is necessarily
conditional on context, as you can see above. Whether you choose the
TERGM or the SAOM, for example, should depend on a number of factors,
including whether the inherent theory built into the SAOM is congruent
with the process that generates the data. For example, are the ties in
your dataset based on actors' outgoing choices, are they strictly
sequential, or could actors send ties to multiple recipients at once,
etc.? Similarly, irrespective of model choice, whether you treat
composition change one way or another depends on the circumstances. I
think Carter, Zack, Martina and other colleagues are paving the way for
some exciting developments, and different people will be using them in
different ways, depending on need and context.

Best regards,

Philip


--
Philip Leifeld
Professor, Department of Government
University of Essex
http://www.philipleifeld.com

On 18/08/2020 08:41, Christian Steglich wrote:

> Hi Holly and all,

>

> if you're prepared to switch to the continuous-time framework of

> stochastic actor-oriented network models, the RSiena software offers a

> solution in which entry and exit times can be used as exogenous

> information:

>

> *

> Huisman, M., & Snijders, T. A. (2003). Statistical analysis of

> longitudinal network data with changing composition. /Sociological

> methods & research/, /32/(2), 253-287.

>

> Interestingly, there is at least one published TERGM paper in which the

> networks over time are assumed to be independent observations of

> changing size, so composition change (in this case: network exit only)

> is implicitly handled as "sampling another independent network of

> different size from the same ERGM parameters":

>

> *

> Czarna, A. Z., Leifeld, P., Śmieja, M., Dufner, M., & Salovey, P.

> (2016). Do narcissism and emotional intelligence win us friends?

> Modeling dynamics of peer popularity using inferential network

> analysis. /Personality and Social Psychology Bulletin/, /42/(11),

> 1588-1599.

>

> I would personally strongly recommend against taking such an approach,

> for various reasons - but then, I am not a TERGM user and do not know

> about common practice in this part of the methods garden. In fact, I'd

> be eager to read any arguments that might justify such an approach.

>

> All the best, Christian

>

>

>

> On 18-8-2020 6:51, Zack Almquist wrote:

>> Hi Holly and Carter,

>>

>>

>> Just to follow Carter's link, since I think my email didn't post.

>>

>> One approach would be to use to dnr package

>> (https://cran.r-project.org/web/packages/dnr/index.html

>> <https://linkprotect.cudasvc.com/url?a=https%3a%2f%2fcran.r-project.org%2fweb%2fpackages%2fdnr%2findex.html&c=E,1,EOxSGnCxGiPtTW3Lo0YKsBNvzW_DCDQW9i1rEcsbuRU84_mmZRGnxjTfIv7ZacOmjtAH_h5Nee1XMIrNJjPH5ILSiVqP-ETXl0D9OPCsZg,,&typo=1>)

>> which is based on

>>

>> Almquist, Zack W., and Carter T. Butts. "Logistic network regression

>> for scalable analysis of networks with joint edge/vertex dynamics."

>> /Sociological methodology/ 44.1 (2014): 273-321.

>>

>> and

>>

>> Mallik, Abhirup, and Zack W. Almquist. "Stable Multiple Time Step

>> Simulation/Prediction From Lagged Dynamic Network Regression Models."

>> /Journal of Computational and Graphical Statistics/ 28.4 (2019): 967-979.

>>

>> Best,

>>

>> Zack

>> ---

>> Zack W. Almquist

>> Assistant Professor

>> Department of Sociology

>> Senior Data Scientist Fellow, eScience Institute

>> University of Washington

>>

>>

>> On Mon, Aug 17, 2020 at 9:46 PM Carter T. Butts <buttsc at uci.edu

>> <mailto:buttsc at uci.edu>> wrote:

>>

>> Hi, Holly -

>>

>> This can certainly be done, although it requires a non-trivial

>> extension of the modeling framework.  One treatment can be found

>> here: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4479214/

>> <https://linkprotect.cudasvc.com/url?a=https%3a%2f%2fwww.ncbi.nlm.nih.gov%2fpmc%2farticles%2fPMC4479214%2f&c=E,1,YvTowsnxQM7shqd8prhLj09-AfKDUumr0PPXoFf36g_A8KrX8x7f_KhBYk6kvdc3DYa48Aj4OD_B1-FCudxySEcxVZUtnoDsIRx8ZFfbkNA2NA,,&typo=1>

>>

>>

>> Hope that helps!

>>

>> -Carter

>>

>> On 8/17/20 12:46 PM, Holly Loncarich wrote:

>>>

>>> Hello,

>>>

>>> I hope that everyone is having a good start to their semester! I

>>> am attempting to use a TERGM over a time period of five years on

>>> a strategic alliance dataset. I was wondering if anyone has any

>>> advice or could point me to any resources on handling node entry

>>> and exit in the TERGM analysis? I am very new to TERGMs and would

>>> appreciate any thoughts or comments.

>>>

>>>

>>> Thank you,

>>>

>>> Holly

>>>

>>> Holly Loncarich

>>>

>>> PhD Candidate

>>>

>>> Walton College of Business

>>>

>>> University of Arkansas

>>>

>>>

>>> _______________________________________________

>>> statnet_help mailing list

>>> statnet_help at u.washington.edu <mailto:statnet_help at u.washington.edu>

>>> http://mailman13.u.washington.edu/mailman/listinfo/statnet_help

>> _______________________________________________

>> statnet_help mailing list

>> statnet_help at u.washington.edu <mailto:statnet_help at u.washington.edu>

>> http://mailman13.u.washington.edu/mailman/listinfo/statnet_help

>>

>>

>> _______________________________________________

>> statnet_help mailing list

>> statnet_help at u.washington.edu

>> http://mailman13.u.washington.edu/mailman/listinfo/statnet_help

>

> --

> ------------------------------------------------------------------------

>

> *Interuniversity Centre for Social Science Theory & Methodology*

> Department of Sociology, Grote Rozenstraat 31, NL-9712 TG GRONINGEN

> https://steglich.gmw.rug.nl

> <https://linkprotect.cudasvc.com/url?a=https%3a%2f%2fsteglich.gmw.rug.nl&c=E,1,WW_zA3CTyRL7ocBgFEboDzH5pjmPLvL6iAPO_6Zj_RtLGWjXbBQe_HW3-4M8V3QyE36vhLjfdcMEES4_Hq3uuYkBqShcdN5BE7upzSqP-J0AjTuNmM6I_z8LBpX-&typo=1>

>

> ------------------------------------------------------------------------



More information about the statnet_help mailing list