Error Using Normalization

Submitted by RaviBot on


I am importing thresholded weighted, undirected matricies (symmetric) into GraphVar from an outside progam. These networks/matricies are all not fragmented (checked using the CheckFrag button). I am randomizing using the random_und_connected function however when I try to calculate cetain graph metrics such as betweenness centralitity, I get an error saying "Normalization error: some metrics derived from random networks result in zero which can not be used for normalization." 

Could someone please advise on what could be the issue? Thanks in advance!

Johann Kruschwitz

Tue, 05/16/2017 - 09:47

the problem you are reporting is rather a graph theoretical question than a normalization issue. Some graph measures can by default reach values of zero - e.g., in betweenness centrality nodes will have a value of zero if they do not participate on any shortest paths (between the other nodes) in a network. Thus, even if your networks are not fragmented, values of zeros are likely for some nodes on the outer boundary of the graph. You can check this if you compute e.g. betweenness cantrality and have a look at the saved results (your workspace/results/interim results ID/GraphVars) - if you load the corresponding file into matlab you will see the respective betweenness centrality of each node per subject. For some metrics you could substitute the random network derived zeros by 1 to make normalization possible - for others normalization by random networks does not make sense (e.g., undirected binary degree: when randomizing an undirected network while perserving the degree distribution).

Here is a link to a coursera slide show that explains this issue wrt betweenness centrality: