help
help > RE: Between group contrasts and graph theory
Jan 27, 2017 02:01 AM | Alfonso Nieto-Castanon - Boston University
RE: Between group contrasts and graph theory
Hi Christina,
Yes, this is a bit of an arbitrary choice, the 0.15 value is chosen by default because that very often corresponds to an approximate "small-world" regime. Leaving the threshold empty in the gui and pressing enter will take a bit but it will construct a plot showing you the global and local efficiency of the networks resulting from using a range of cost threshold values. This plot will also show you for comparison the expected global&local efficiency of a random graph and a lattice. Trying to find a reasonable threshold or threshold range that shows relatively high global efficiency compared to that of a lattice, and relatively high local efficiency compared to that of a random graph is what leads, often, to a threshold value around 0.15 (but your data may vary). Also keep in mind that you may also enter multiple values in the cost threshold field, and that, as you mention, will use those multiple thresholds to compute different networks, and the graph measure of interest (e.g. global efficiency) will then be aggregated across all those different networks (making the results a bit more robust to the specific choice of cost threshold).
Hope this helps
Alfonso
Originally posted by Christina van den Brink:
Yes, this is a bit of an arbitrary choice, the 0.15 value is chosen by default because that very often corresponds to an approximate "small-world" regime. Leaving the threshold empty in the gui and pressing enter will take a bit but it will construct a plot showing you the global and local efficiency of the networks resulting from using a range of cost threshold values. This plot will also show you for comparison the expected global&local efficiency of a random graph and a lattice. Trying to find a reasonable threshold or threshold range that shows relatively high global efficiency compared to that of a lattice, and relatively high local efficiency compared to that of a random graph is what leads, often, to a threshold value around 0.15 (but your data may vary). Also keep in mind that you may also enter multiple values in the cost threshold field, and that, as you mention, will use those multiple thresholds to compute different networks, and the graph measure of interest (e.g. global efficiency) will then be aggregated across all those different networks (making the results a bit more robust to the specific choice of cost threshold).
Hope this helps
Alfonso
Originally posted by Christina van den Brink:
Hi
Alfonso,
Thank you for that! I do have a follow-up question. I've noted that the default threshold for graph theory metrics is 0.15. Is there a specific reason that the top 15% of edges are selected? I work with aging data, and I've noted that many papers typically use a range of thresholds to accommodate for variability in aging networks. Would you have any suggestions when it comes to determining thresholds?
Best,
Christina
Originally posted by Alfonso Nieto-Castanon:
Thank you for that! I do have a follow-up question. I've noted that the default threshold for graph theory metrics is 0.15. Is there a specific reason that the top 15% of edges are selected? I work with aging data, and I've noted that many papers typically use a range of thresholds to accommodate for variability in aging networks. Would you have any suggestions when it comes to determining thresholds?
Best,
Christina
Originally posted by Alfonso Nieto-Castanon:
Hi
Christina,
In order to look at potential between-group differences you need to select the four groups and enter a between-subjects contrast of the form [1 -1 0 0; 0 1 -1 0; 0 0 1 -1] (this will appear labeled as 'any differences between ...' in the contrast-help list). Then, after defining the network of ROIs and measures of interest, the first line in the results table (labeled 'network') will tell you whether there are any differences between the groups in your choice of network metric at the network-level (e.g. is network global efficiency different between the groups?), and the rest of the lines will tell you whether there are any differences between the groups in the same metric at the individual node/ROI level (e.g. is the global efficiency of the MPFC node different between the groups?)
Hope this helps
Alfonso
Originally posted by Christina van den Brink:
In order to look at potential between-group differences you need to select the four groups and enter a between-subjects contrast of the form [1 -1 0 0; 0 1 -1 0; 0 0 1 -1] (this will appear labeled as 'any differences between ...' in the contrast-help list). Then, after defining the network of ROIs and measures of interest, the first line in the results table (labeled 'network') will tell you whether there are any differences between the groups in your choice of network metric at the network-level (e.g. is network global efficiency different between the groups?), and the rest of the lines will tell you whether there are any differences between the groups in the same metric at the individual node/ROI level (e.g. is the global efficiency of the MPFC node different between the groups?)
Hope this helps
Alfonso
Originally posted by Christina van den Brink:
Hello Conn community,
I am setting up the contrast portion of my analysis and wanted to confirm that I'm on the right track. In my analysis, I have four levels of my variable of interest (body mass index). The goal is to examine local and global efficiency within networks of selected ROIs, and detect between-group differences therein.
What I would like to see is how group membership affects these properties at the network level and then work backwards identifying group differences at the nodal level. At this point, I've set up an initial contrast [1 0 0 0; 0 1 0 0; 0 0 1 0; 0 0 0 1] which is looking for any effect of group membership on these properties. I've attached an image of the output. My interpretation of the output is that there is a significant effect of group membership on local efficiency at the network and nodal level. But I can't decipher how to obtain further information as to where these differences lie in terms of membership or in what direction. Is there an omnibus way to do this? Or should I structure specific contrasts?
Many thanks for your time and help!
I am setting up the contrast portion of my analysis and wanted to confirm that I'm on the right track. In my analysis, I have four levels of my variable of interest (body mass index). The goal is to examine local and global efficiency within networks of selected ROIs, and detect between-group differences therein.
What I would like to see is how group membership affects these properties at the network level and then work backwards identifying group differences at the nodal level. At this point, I've set up an initial contrast [1 0 0 0; 0 1 0 0; 0 0 1 0; 0 0 0 1] which is looking for any effect of group membership on these properties. I've attached an image of the output. My interpretation of the output is that there is a significant effect of group membership on local efficiency at the network and nodal level. But I can't decipher how to obtain further information as to where these differences lie in terms of membership or in what direction. Is there an omnibus way to do this? Or should I structure specific contrasts?
Many thanks for your time and help!
Threaded View
| Title | Author | Date |
|---|---|---|
| Christina van den Brink | Jan 25, 2017 | |
| Alfonso Nieto-Castanon | Jan 27, 2017 | |
| Christina van den Brink | Jan 27, 2017 | |
| Alfonso Nieto-Castanon | Jan 27, 2017 | |