SOCNET is a service of INSNA, the professional association for social
network researchers (http://www.insna.org). To unsubscribe, send
an email message to
I write with a few questions about network regression and hope someone might be able to offer some advice.
I collected data about which environmental restoration groups work together in a region. One of the variables I collected was how productive the felt each partnership was for meeting their restoration goals. This is a dyadic measure and I collected this information as ordinal data on a 5 point scale. I want to understand if productivity is related to other dyadic variables that I collected (e.g. frequency of interaction, age of partnership, being the of same governance type (e.g. local vs state gov)). All my data is ordinal or nominal data so I am not sure if I should use regular QAP OLS regression (which can be done in UCINET) or if I need to do a QAP logistic regression, in which case I would have to use R-sna package. I have come upon internet examples where people used QAP OLS regression with ordinal data, but normally one would do logistic regression with ordinal data so I wanted to get some feedback about what others have done and why.
I am also curious to see if the productivity scores (dyadic, basically a tie strength) are related to in/out degree and centrality. So in this case I have tie strength as dependent and node attributes as independent. How would I test this? Given that my node level independent variable is interval or ratio, could I use the Moran's I and Geary's C tests in UCINET (tools > mixed dyadic/nodal > continuous > moran/geary statistics)?
Is it possible to include the node attributes (e.g. degree) and tie values (e.g. interaction frequency) in the same list of independent variables?
I have some missing attribute data for some ties. That is, I know two nodes are connected, but dont know information for the frequency of interaction or the productivity because the survey participant skipped some questions. Can I include missing data in the QAP models, or do I have to throw out all relationships for which I do not have information for all independent and the dependent variables?
If anyone can offer advice with these questions I would be grateful.
-- Jesse Sayles
School of Geographical Sciences & Urban Planning
Arizona State University