It seems that in the first case you would want to follow the Snijders and Baerveldt approach. See:
Snijders, Tom A.B, and Baerveldt, Chris, A Multilevel Network Study of the Effects of Delinquent Behavior on Friendship Evolution. Journal of Mathematical Sociology, 27 (2003), 123-151.
Also see the siena08 function.
The second approach seems to require you to restrict the permutation options, so that you cannot have ties between people of different groups. In ERGM (p*) and SIENA this is done by using structural zero’s (ties that are required to be 0 for relations between people of different groups). For QAP you might want to contact David Dekker, who did a workshop on QAP using UCINET and R at the recent Sunbelt.
Hope this helps,
Dr. Filip Agneessens
Senior Lecturer in Organizational Behaviour and Human Resource Management
Surrey Business School, University of Surrey
From: Social Networks Discussion Forum [mailto:[log in to unmask]]
On Behalf Of Crawford, Eean R
Sent: 26 August 2015 16:40
To: [log in to unmask]
Subject: [SOCNET] QAP on multiple independent networks with same relations
***** To join INSNA, visit http://www.insna.org *****
I believe a question like this may have surfaced on SOCNET before, but I couldn't find it in my search of the archives (either SOCNET's or my email archives). Any guidance or literature you can point me toward will be most welcome.
I have network data on 30 different teams at two points in time. The teams are quite large (average team size is 18 people, range is 16-20, total number of individual is 540), and there are no ties between teams. I want to know if ties between team members collected during week 1 (call them X ties) predict ties between those same team members collected seven weeks later (call them Y ties).
I know how to do QAP testing for the association between ties if I were only doing it on a single team. And most of the guidance I have read on QAP testing only illustrates the procedure on a single set of network actors. My question is, what is the appropriate way to do the same QAP testing on 30 independent teams? What do you do when you have the same network relations collected on 30 independent sets of actors? Here are two alternatives I've considered:
I conduct 30 separate QAP tests and then average the results from the tests in a sort of mini-meta-analysis.
QAP Team 1 X with Team 1 Y (matrices are 16 x 16)
QAP Team 2 X with Team 2 Y (matrices are 18 x 18)
QAP Team 3 X with Team 3 Y (matrices are 20 x 20)
QAP Team 30 X with Team 30 Y (matrices are 16 x 16)
If I do this, would I weight the individual team correlations by the network size? What would I do about the p-values (proportion of randomly permuted correlations that were as large) from the individual QAP correlations - does it even make sense to average those?
I create the union of all 30 X matrices, the union of all 30 Y matrices. The union matrices are 540 x 540, which is the total number of individuals across all teams, and the ties within teams are arranged in blocks down the diagonal. The ties in off-diagonal blocks (between teams) are empty represented as missing values (not zeros). I then do a single QAP test on the Union X with Union Y.
QAP Union X with Union Y (matrices are 540 x 540).
If I do this, am I biasing the QAP because its permutations include ties (between teams) that couldn't possibly have existed?
Thanks for helping me think through this.
Eean Crawford, Ph.D.
Assistant Professor, Dept. of Management & Organizations
Tippie College of Business
University of Iowa
W376 John Pappajohn Business Bldg.
Iowa City, IA 52242-1994
Ph: (319) 335-2884
Fx: (319) 335-1956
_____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message.