Dear NBS experts,

Is it possible to feed a matrix containing negative values into NBS? Or should all values be absolute?

thank you in advance,

kind regards,
Hans van der Horn

Hi Hans,

In short, yes, a connectivity matrix comprising negative values is fine.

It is also fine to take the absolute value of all values in the connectivity matrices.

Both these approaches are common and acceptable. They do however have different interpretations.

Andrew

Originally posted by Hans van der Horn:

Dear Andrew,

Thank you very much for your reply. Could you perhaps elaborate a bit more on the interpretation differences?
I can imagine that differences between groups can occur with running the analysis on absolute values, but may be absent with running it on positive and negative values. That is, negative values converted into positive values may cause group differences which are not actually true. Also I can imagine that finding differences on group level with absolute values, has to be interpreted in the light of the actual correlation values.

Thanks again!

kind regards,
Hans van der Horn

Hi Hans,

Yes - you are right - whether or not negative connectivity values are forced to be positive with their absolute value can affect the set of connections for which the null is rejected. 

Negative functional connections can be an artifact of global signal regression (GSR). Avoiding GSR alleviates the issue to an extent, although GSR might be useful for head motion correction and removal of other noise sources. In any case, how to deal with negative connections remains an open issue. Do a search in the journal NeuroImage on "global signal regression" or "negative weights" for further details. Also see Rubinov's 2011 paper, "Weight-conserving characterization of complex functional brain networks".

I'd suggest trying both approaches - with and without absolute values. 

Andrew

Originally posted by Hans van der Horn:

Thanks Andrew!:)