I am a physician and visiting research scientist at yale and together with my coauthors have published several studies in the field of psychology through Network Modeling and related analyses using r packages like bootnet, qgraph, NetworkComparisonTest, etc. I thought it might be a good idea to discuss different subjects related to this field, in a group. I am also interested in collaboration with other teams working in the field. Let's share our ideas, questions and suggestions in this group.
Article Network approach in health behavior research: how can we exp...
If I understand your concept correctly, you create networks where the nodes are the symptoms and their effects, and edges link the symptoms with their corresponding effects, correct? So are these multimodal networks?
Exactly. I will attach my latest publication, so that you could get a more precise sense of the methodology.
@annamalgorzatakaminska , one thing to mention is that the graphs are a bit different than MMNs and are Gaussian Graphical Models, where all edges represent the regularized Partial Correlation Coefficients matrix so an edge between two nodes is slightly smaller in weight than the partial correlation coefficient, additionally, the thinnest edges are deleted during the regularization process. The goal is to sparse the network and delete spurious coefficient produced due to multiple testing during generation of the partial correlation matrix, while minimizing the effect on overall model accuracy. These regularized network models have advantages and disadvantages over the non-regularized networks
Hi Farhad,
I'm not quite sure if it's relevant to what you're working on or not. But anyway, if you made any network with the purpose of finding the right candidates, you may consider using the IVI, an integrative algorithm, for finding the most influential nodes within the network! It is included as a function in the Influential R package (https://cran.r-project.org/web/packages/influential/index.html).
You may also read the corresponding paper at https://www.researchgate.net/publication/342374701_Integrated_Value_of_Influence_An_Integrative_Method_for_the_Identification_of_the_Most_Influential_Nodes_within_Networks
Hi Abbas Salavaty ,
I read the article and checked the package. I am impressed by the work. It definitely helps. I also have a suggestion. I guess the functions would be easier to be implemented by the user, if the arguments are extracted from the input, inside the function, from the main input, rather than the user specifying them in the list of arguments. For example in the ivi function, the once the weight argument is extracted from the igraph input, if turned into an adjacency matrix, the direct/indirect nature of the graph could be understood based on the symmetricity of the matrix, so the mode argument could be set within, as well as the loop argument.
Thanks a lot for sharing your work and introducing the package.
Best,
Farhad
Hi Farhad Montazeri
Thanks for your suggestions. I appreciate it. However, most of the arguments of the "ivi" function have their default values and are not required to be specified in most cases. Also, the arguments give the opportunity to calculate the IVI values of a directed graph, while considering it as undirected, for instance!
Abbas Salavaty ,
Makes sense. Thanks for letting me know.
I hope followers of this discussion forum check your study and package,
which address an interesting topic in network analysis. Let's keep in touch!
Best Wishes,
Farhad
Farhad Montazeri
I'm happy it could help. Thanks for promoting my research and the influential R package.
We'll be in touch,
Adrian (Abbas)
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R