Hi!
I'm trying to run a basic seed-level connectivity analysis with data that have already been pre-processed (before importing into CONN) using the list of steps below. However, the data have not been pre-smoothed. Thus I am running the smoothing pre-processing stage in CONN with the default 8mm kernel. Then, because we already ran compcor etc., I am skipping all the denoising steps (except the bandpass filtering at the default values, because the data were not band-passed). Yet, even though we already ran compcor etc before importing into CONN, the histograms in the denoising step show a definite bias to the right, at about .2-.4. And the correlations in the first level analysis look too strong / widespread. When I load up the unsmoothed data, the denoising histograms don't have this bias to the right and appear much more centered. But then the data aren't smoothed so it's a bit of a mess and we find no effects. Back to the smoothed data, If I add the CSF and White matter as confounds in the denoising step, the histograms look much more centered. However, it seems inappropriate to rerun the compcor denoising since we already did it before importing the functional data into CONN. When I tried denoising the CSF and WM again in CONN, we didn't really find any of the effects we were expecting, and I was worried that perhaps by double denoising—once before importing in CONN and once in CONN after smoothing—we somehow got rid of valuable information in our data.
Any ideas about how to proceed? 
Thank you so much!
Michael

Here's the data preprocessing BEFORE importing into CONN:
Functional images were pre-processed using MATLAB 2012 (The Mathworks Inc., Natick, MA, USA), SPM12 (Statistical parametric mapping software, SPM; Wellcome Department of Imaging Neuroscience, London, UK; http://www.fil.ion.ucl.ac.uk) and DPABI v2.1 (toolbox for Data Processing & Analysis for Brain Imaging; http://rfmri.org/dpabi; Yan and Zang, 2010; Yan et al., 2016).
Reorientate Fun* and T1* to oblique space
Remove first 5 volumes
Slice timing (TR = 2.3 sec; nslice = 37; slice acquired [1:2:37,2:2:36]; reference slice = 19)
Realign (micromovements estimated as FD calculated according to Power et al,. 2012, FD_Power >= 0.5 is considered bad time points; a subject is excluded if his or her bad time point rate exceeds 15% or any of the 6 rigid body head movement parameter exceeds 3 mm or degree)
T1 coregistered to Fun
T1 Segmentation using DARTEL
Normalise by using DARTEL (bounding box [-90 -126 -72;90 90 108]; voxel size [3 3 3])
Nuisance covariate regression
polynomial trend = 1; nuisance regressors includes:
6 rigid body head movement parameters
first 5 principal components from WM & CSF signal according to the CompCor algorithm (component based noise correction method also used in Conn & C-PAC; Behzadi et al., 2007)