***** To join INSNA, visit http://www.insna.org *****
I am fairly new to the field of SN, and have done some
reading of Wasserman and Faust, as well as going
through the archives of this listserv and the yahoo
group. But i still have a question:
I am interested in modeling help seeking behavior
(related to homework) among a group of students. My
interests are in comparing the individual and
structural influences on (1) who seeks, and (2) who is
sought for homework related help.
My covariates are:
-- individual factors (gender, academic ranking,
degree/betweeness centrality scores for each node)
-- dyadic factors (frequency of communication,
-- structural factors (gender, academic ranking, being
members of same structural equivalence block,
Originally, i estimated two logistic regression models
-one predicting help seeking and the other being
consulted for help.
My readings indicate that this model could be wrong
given the non-independence of observations, and
implied correlation of error terms. So, I understand i
can use the QAP procedure in UCINET.
How do i go about this, so that i am still able to
gauge the relative influence of individual and
structural effects, while controlling of the effects
of other covariates? I was thinking of two
1. estimate my first model using OLS logistic
regression, including only the individual and maybe
the dyadic level predictors (but wouldn't the presence
of the dyadic variables - e.g., Xs reported frequency
of communication with Y or Xs estimation of emotional
closeness with Y) lead to the non-independence issues?
2. estimate the structural side of the model using QAP
- basically, take the original directed matrix (who do
you go to for help), take its transpose to give me
(who is sought for help), and then regress this on my
homophily, proximity matrices. I can even include the
dyadic variables here, by includding the communication
and emotional closeness matrices as additional
3. The issue i have with the second model is that the
predictors would include both symmetric (homophily,
proximity) and assymetric (communication, emotional
closeness) matrices. And given that my dependent
variables are intrinsically asymmetric, wouldn't there
be some measurement error? To reduce/avoid this,
would it be better to first binarize and symmetrize
the communication and emotional closeness variables
(thus losing a lot of information) before including
them in the model?
4. In otherwords, can my predictos include both
binary, valued, symmetrical and asymetrical relations?
I appreciate your assistance..
Do you Yahoo!?
Yahoo! Mail - 250MB free storage. Do more. Manage less.
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.