Hi Fred,

There are three ways in CONN for removing one or several subjects from a particular second-level analysis (e.g. correlation of connectivity with IQ with some missing-data in the IQ variable):

1) first define a new set of second-level covariates where the missing subject data is filled with 0's in all of the covariates that are relevant to your analysis (e.g. create a new 'AllSubjects_valid' covariate and a new 'IQ_valid' covariate, where the values for the missing subject have been filled with 0). Then perform the standard second-level analysis using these new covariates (e.g. select 'AllSubjects_valid' and 'IQ_valid' and enter a contrast [0 1])

2) first define a new set of second-level covariates, one for each missing-data subject, dummy coding the corresponding subject(s) (e.g. create a new 'Subject1_missing' covariate that contains all 0's except for the missing subject where you enter a 1). Then perform the standard second-level analysis using the typical subject effects PLUS any newly create 'Subject_mising' covariate(s), for which you enter a 0 in the corresponding contrast definition value(s) (e.g. select 'AllSubjects', 'IQ', and 'Subject1_missing', and enter a contrast [0 1 0])

3) in release 15 you may also use a third approach where you can simply enter NaN for any values that are missing in your second-level covariate definitions (e.g. in your 'IQ' covariate simply enter NaN as the value for the missing-data subject(s)). Then perform the standard second-level analysis using the typical subject effects (e.g. select 'AllSubjects', and 'IQ', and enter a contrast [0 1])

All of these approaches are exactly equivalent, they will lead to exactly the same test, statistics, p-values, degrees of freedom, etc. You may simply choose the one that you find simpler to implement for your particular case. The reason why approach (1) works even for continuous variables like IQ or DMI, which may contain actually valid/meaningful 0 values, is that CONN will only remove subjects from consideration in your second-level analysis if they contain 0's in all of the included between-subject effects. An actual missing-DMI subject will contain 0's in both the 'AllSubjects_valid' as well as the 'DMI_valid' variables, while a valid-DMI subject may contain 0's in the 'DMI_valid' variable, but it will always contain 1's in the 'AllSubjects_valid' variable so they will not be removed from the analysis.

Hope this helps clarify
Alfonso

ps. regarding your last "I dont' want to remove/exclude a whole participant from an analysis" remark, I am not sure if I am underestanding or addressing that question correctly, could please elaborate what you mean there?

Originally posted by Fred Uquillas: