Hi, I¡¯m afraid I can¡¯t understand your question. I get that you have done a number of correlations, and found that most of the negative correlations are not significant (all but the very first one). You find that suspicious, but those are also the correlations based on very small N¡¯s ¨C usually just 3 nodes. The correlations based on more nodes (eg. 12), are still not significant because the correlation coefficients themselves are small. This strikes me as normal, but is difficult to say more as I really don¡¯t understand you data. E.g., why are there 31,718 rows with no data, and another 7,629 with values of NA?
?
I you would like to talk further, email me at sborgatti@....
I calculated how much type I relations correlate (QAP Correlation) with type II relations for the ego network, of course, noting that the type I egonet should contain the same nodes as the type II egonet. To do this, I performed the union of the nodes of each egonet to ensure that the same nodes would be present in both egonets. The results are in the attached csv. The egonets are directed and valued. I expected inverse CorrCoefs; however, all calculations related to the inverse presented a p-value > 0.05. If it had occurred with some cases, I would understand. But all?
Question: It seems that QAPCorrelation may have been designed for direct correlations. And that suits me. If true, then I should have symmetrized (and perhaps dichotomized) the networks. Is that correct?
