Hi Andrew,

Recently, I have come across a problem when using NBS. I compared the networks between two groups and set the thresholds from 2.2 to 3.0, separately. All these comparisons found significant results. But among these results, there was a strange thing. For example, when the parameters set as 2.2-2.5, the corresponding significant results consistently located within A, B, C and D edges. However, once the parameter values set above 2.5, the corresponding significant results were in F, G and T edges. I am very confused with these results why when the parameter values set above 2.5, the significant results were not consistent with those parameters setting below 2.5? Were these results reasonable? Why did appear such phenomenon? Would you like to give me some suggestions to understand it? Thank you very much in advance!

Best regards,
Jiaolong

Hi Jiaolong,
What you describe is possible. For a threshold below 2.5, the edges F, G and T only comprise a component of size 3, which was probably too small to reach significance. However, once you increased the threshold above 2.5, a component of size 3 was sufficient to reach significance. But the test statistic at edges A, B, C and D might not have exceeded 2.5 and thus this component disappeared. 
Andrew

Originally posted by jiaolong qin:

Hi Andrew,

Thank you so much for your detailed explanation and I see it. BTW, If I need to choose one result from these two different results, which factors should I consider? If you figured out such results, which result would you select? Thanks again. 

Best regards,
Jiaolong
Originally posted by Andrew Zalesky:

It's hard for me to comment because I am not familiar with your study. However, as a general rule, I suggest using the highest possible primary threshold, but also discussing the result given by the lower threshold. I hope that helps.

Andrew

Originally posted by jiaolong qin:

Hi Andrew,

Thank you very much for your time and kindly help.

Best regards,
Jiaolong

Originally posted by Andrew Zalesky:

Hi Andrew,

My topic is related to this previous discussion in the forum. Everything works perfectly fine for my study but I just want to understand more.
I wonder why when I decreased the initial T threshold (initially was set up between 3.5-4, giving me beautiful and very consistent results, from a bigger subnetwork including more edges to a more resume subnetwork comprising only the stronger connected edges). Now I decreased the T threshold to 2.5 and the subnetwork previously obtained completely disappears, setting the component size for extend or intensity. Why?
Thanks a lot in advance.

Lorna.

Hi Lorna,

There is no right or wrong threshold. Some details about the choice of threshold are provided in the manual.

A t-threshold of 3.5-4 is quite conservative. If you have consistent findings within this range, I suggest that you stick with it.

If you lower the threshold to 2 and the sub-network disappears, this simply means that the sub-network extent or intensity does not exceed the null.

Andrew
Originally posted by Lorna Garcia Penton:

Thanks a lot, Andrew!