[Rsiena-help] checking for correctly predicted network evolution

[Tom Snijders]

Dear Mark,

By the way, this type of question is more appropriate for the list
http://groups.yahoo.com/groups/stocnet/
than for this one - this one focuses on coding issues.

Predicting binary outcomes is very difficult in general, and I am not a fan of judging statistical models on their ability in this respect.
For network models the situation is even more complicated, because here the dependence between tie variables is an essential part of the model, and not captured by counting correct predictions. For example, a network model will predict that there are a lot of closed triads, but a longitudinal network model will not predict where they are (unless some open triads are available already).
This is the reason why sienaGOF() focuses on occurrence of patterns (e.g., goodness of fit of the triad census) rather than individual tie predictions.

Best wishes,
Tom

================================================================
Tom A.B. Snijders
Professor of Statistics in the Social Sciences
Department of Politics and Department of Statistics
Nuffield College
University of Oxford
tel. +44-01865-278599

From: rsiena-help-bounces at lists.r-forge.r-project.org [mailto:rsiena-help-bounces at lists.r-forge.r-project.org] On Behalf Of Mark Manger
Sent: 02 July 2014 22:06
To: rsiena-help at lists.r-forge.r-project.org
Subject: [Rsiena-help] checking for correctly predicted network evolution

Dear fellow RSiena users,

I tend to apply RSiena to problems that in Political Science are usually treated with binary dependent variable models, most commonly logit models.

With logit models, because it is relatively simple, people often calculate a "percentage correctly predicted" of outcomes, and then perhaps list which specific outcomes were over- or underpredicted. This is generally (perhaps wrongly) considered as a good check of the explanatory power of a model, and frequently requested by reviewers.

I'm wondering if it would be possible to do the same in RSiena - predict (or rather simulate the networks given my estimated parameters), compare with the actually observed networks, and calculate a percentage correctly predicted in each wave?

-Mark

Mark S. Manger
Assistant Professor of Political Economy and Global Affairs
Munk School of Global Affairs & Department of Political Science | University of Toronto
Observatory Site | 315 Bloor Street West | Room 212
Toronto, ON   M5S 0A7
Phone: 416-946-8927 | Fax: 416-946-8877
mark.manger at utoronto.ca<mailto:mark.manger at utoronto.ca>
www.munkschool.utoronto.ca/mga<http://www.munkschool.utoronto.ca/mga>

JOIN THE GLOBAL CONVERSATION



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/rsiena-help/attachments/20140707/bc2b53a6/attachment.html>