Dear all,

I have a weighted correlation matrix for which I need to find various metrics like clustering coefficient, path length etc based on specific threshold values. Since the matrix has all non-zero elements, please suggest me on how to find threshold based on sparsity?

Hi,
There is not a golden rule for thresholding, but for the path-length is reasonable to remove at least the negative distances. My suggestion is to test your sample inferences at least in several sparsity levels to evaluate the robustness of your results. (e.g., from 5-40% of sparsity in steps of 5%).
For more info, this is a good start for the topic: https://youtu.be/MZEQR7gKfNE
Best,
Zeus.

Thanks Zeus.. I observed that sparsity and density are exactly the opposite things,  considered in many studies. A Sparsity of 20% ignores 20 % of the connections and considers top 80% of the connections, while a density of 20% considers 20% and ignores the rest. Hope I understood it right.

Originally posted by TANMAYEE SAMANTARAY:
Yes, you are right. I meant density when I was talking about sparsity. Sorry. I'd apply different levels of density (such as: https://doi.org/10.1523/JNEUROSCI.3272-1... or https://doi.org/10.1016/j.neuroimage.202...).

Thanks!